M

June 6, 2025 08:58 PM GMT

Software | North America Morgan Stanley & Co. LLC

Foundation

Sanjit K Singh
Equity Analyst

AI Agents Knocking at the Door [email protected]

Keith Weiss, CFA

+1 415 576-2060

Equity Analyst
[email protected] +1 212 761-4149

The growing power of deep reasoning models pave the way for Brian Nowak, CFA
Equity Analyst
a new wave of Agentic AI applications able to automate [email protected] +1 212 761-3365

broadening business functions, unlocking tremendous value. The Josh Baer, CFA
Equity Analyst
customer journey to Agentic AI will take time and software [email protected] +1 212 761-4223

players will have to evolve business + pricing models to thrive Elizabeth Porter, CFA
Equity Analyst
[email protected] +1 212 761-3632

Key Takeaways Chris Quintero

Equity Analyst

Over decades software was developed based on the assumption that humans [email protected] +1 212 761-1686

play the central role in executing tasks, workflows and business processes. Oscar R Saavedra
Research Associate
[email protected] +1 212 761-0827
Agentic computing removes this assumption, replacing it with the idea that
Agents increasingly become the entity executing tasks and making decisions

The value creation opportunity is enormous as Agentic applications penetrate the

$630+ billion application software market over time

However, the transition to agentic will come with change – redefining how apps
Software
are architected and requiring new business and pricing models North America
Industry View Attractive
Software companies well positioned to benefit include: AMZN, CRM, CYBR,
GOOG, HUBS, MSFT, NET, NOW, OKTA, OS, PLTR, SAIL, SNOW, & TEAM

For Deep Dive in the Agentic Computing Opportunity Please see our Agents
Knocking on the Door Presentation attached to this report

Bringing Agency to Software. Over the last several decades, software has been
developed based on the assumption that humans play the central role in executing
tasks, workflows and processes. The expanding capabilities of Agentic AI removes
this long-held assumption, replacing it with a framework where AI Agents
increasingly become the entity executing tasks and making decisions. The coalescing
forces driving this shift – the increasing power of deep reasoning models, the ability
to have a semantic understanding of data and the declining price of intelligence –
create a substantial opportunity for enterprises to automate entire business
functions and processes, unlocking enormous value via digital labor replacement.
However, investors should view the march towards agentic computing in the
enterprise as an evolutionary journey rather than a revolutionary sprint. The
inherently indeterminate nature of foundational models, a proliferation of
application and data silos built up over decades, persistent challenges with data
quality, an ever-growing list of security concerns and a lack of standardization with
Morgan Stanley does and seeks to do business with
respect to governance models and application frameworks present significant companies covered in Morgan Stanley Research. As a result,
obstacles to deploying Agentic AI applications – challenges the industry will have to investors should be aware that the firm may have a conflict of
interest that could affect the objectivity of Morgan Stanley
solve over time. Research. Investors should consider Morgan Stanley
Research as only a single factor in making their investment
Bracing for Change. Taking in proper historical context, Agentic computing decision.
For analyst certification and other important disclosures,
represents the next progression of a long-standing evolution towards greater
refer to the Disclosure Section, located at the end of this
abstraction in software. At a basic level, an AI Agent is a software program defined report.
M
Foundation

by key attributes including the ability to: 1) understand and interact with its
environment (receive inputs), 2) collect and connect to data, 3) take action
autonomously without using pre-determined rules and 4) learn from its experience
over time. Like past waves, the transition towards AI Agents will unleash greater
automation, productivity and efficiency, but also force the software industry to
redefine how applications are built and architected, as well as evolve business
models including instituting new approaches to pricing.

System of Record vs. System of Engagement – A Framework for Organizing the

Agent Landscape. The emerging Agentic AI landscape can be organized through a
bifurcated framework distinguishing between System of Record (SoR) Agents and
System of Engagement (SoE) Agents, each addressing distinct automation
paradigms and market opportunities.

• SoR Agents function as domain-specific, functional automation solutions

attempting to directly replace human roles in well-defined processes such as
coding, customer service, sales lead generation, business intelligence, etc.
Examples include GitHub Copilot, Salesforce Agentforce, and HubSpot
Prospecting Agent. These Agents target a narrower market opportunity,
which we closely align with IDC's labor-centric automation market segment,
representing a $16 billion opportunity today projected to reach $33 billion by
2028.
• SoE Agents operate as supervisory orchestrators focused on workflow
and process automation across multiple systems, delivering value through
improved system utilization and indirect time saving rather than direct labor
replacement. Examples include ServiceNow Now Assist, GitLab Workflows,
and Atlassian Rovo. These Agents target a broader market opportunity,
which we closely align with IDC's system-centric and decision-centric
automation, and business value engineering market segments, representing a
$36 billion opportunity today projected to reach $69 billion by 2028.

Exhibit 1: Framework for Organizing the Agent Landscape – System of Record

style Agents vs System of Engagement style Agents.

Source: Morgan Stanley Research

Where Are We on the Path to Agentic AI? Our industry conversations on

2
M
Foundation

production deployments of AI Agents have been mixed – customers report that,

when "working", results are impressive and easily exceed ROI mandates to warrant
further investment. In a significant portion of cases, however, reliability, accuracy
and precision have been disappointing. While this may suggest that the AI reasoning
models are limited in their ability to handle complex tasks, we stress that the
situation is fluid and major advancements in reasoning capabilities appear to be
unfolding in a time frame measured in weeks. Underscoring this point, according to
AI research firm METR, AI agents have doubled the length of tasks they can
automate every 7 months for the last 6 years.

To contextualize where the market stands in terms of agent-based automation

capabilities, AI agent provider Sema4.ai, uses a 5 level framework. Currently, the
market is at level 3 on the path to fully autonomous Agentic AI. This represents the
first level of constrained autonomy – moving beyond rule-based automation and
task-specific co-pilots. At this stage, AI Agents can create, execute, and adapt plans
based on feedback while handling complex tasks through multiple reasoning cycles.

For example, Salesforce's Agentforce agents can reconcile a 100-page invoice against
internal systems with human-like reasoning. This represents a substantial leap from
the lower levels of rule-based automation/robotic process automation (RPA) (level
0), basic AI-enhanced process automation (level 1), and task-specific AI assistants/
copilots (level 2). However, the state of the market remains below level 4
capabilities – systems that can self-improve and modify their own instructions, and
well below level 5 capabilities – fully autonomous AGI-type systems that are
capable of independent reasoning and creativity without oversight. This progression
indicates that while we have made significant progress in autonomous planning and
execution capabilities, human oversight is still necessary in most cases.

Exhibit 2: Length of Tasks That Generalist Frontier Model Agents Can Complete
Autonomously Has Been Doubling Approx. Every 7 Months for the Last 6 Years

Source: Model Evaluation & Threat Research (METR) “Measuring AI Ability to Complete Long Tasks”

How Big is the Opportunity? We evaluate the addressable opportunity for agentic
computing across three distinct lenses that reflect varying degrees of market
penetration and technological maturity. A narrow view, focused on AI-powered
automation, represents a $6 billion opportunity today with projected growth to $20

Morgan Stanley Research 3

M
Foundation

billion by 2028. A wider lens encompassing the broader process automation market
represents a $52 billion opportunity today with projected growth to $102 billion by
2028. An expansive view of the opportunity sees Agentic AI applications penetrating
the entirety of the application software market over time, which represented a $637
billion opportunity today.

Exhibit 3: Expansive View on the Agentic AI Market Points to a $52B Opportunity

Today, Expected to Reach $102B by 2028 (+26% CAGR)

Source: IDC Research

Where Will Value Accrue Across the Stack? Looking across the different categories
of the Agentic AI ecosystem, we see the highest value capture potential in the
following areas.

1. Hyperscalers/AI infrastructure: AI Infrastructure providers have a

significant opportunity in providing the compute to enable the real-time
inference that deep reasoning agents require. This gives leaders in this
category license to not only provide the infrastructure but also the data
retrieval, security, governance and policy enforcement capabilities required
to operate,manage and secure agentic applications.
2. AI Model providers: AI model providers extend the frontier around
reasoning and intelligence that will make AI Agents more capable over time.
This positioning provides the opportunity to extend into areas such as
Agentic frameworks, data processing and integration so as to evolve into
modern application platforms.
3. Security & Governance: We see a significant monetization opportunity in
making Agentic applications more trustworthy, reliable, accurate and secure.
Agent architectures create a large surface area for cyber attacks including
data exfiltration, supply chain attacks, prompt injection and open source
contamination.
4. Data Infrastructure: For AI Agents to be useful, they need governed access
to the most important data sources in real-time. We see two key
opportunities for data infrastructure providers. First, in helping make
Agentic architectures possible by modernizing a customer's data estate and
improving data quality. Second, by enabling fast, reliable and accurate
information retrieval at runtime.
5. Workflow Automation: With AI Agents embedded in workflow automation

4
M
Foundation

platforms, the unstructured components of common business processes can

be more fully automated. As workflow automation platforms span across
key enterprise systems, they become well-positioned to serve as an
orchestration/management layer for both 1st and 3rd party agents.
6. AI Lifecycle Platforms: Agents within AI lifecycle platforms automate the
workflows and tasks associated with developing and deploying AI
applications including building data pipelines, integrating enterprise data
with LLMs, building and evaluating models and visualizing data.
Furthermore, Agents can be embedded into external applications to enhance
customer engagement and to enable highly interactive experiences.

Exhibit 4: Most Value Likely Accrues to the Hyperscalers and AI Infrastructure

Layer as AI Agent Deployments Likely Result in Significant Utilization of Significant
Compute Resources

Source: Morgan Stanley Research

Which Companies are Poised to Benefit? In evaluating how software companies

are positioned for the agentic opportunity, we organize the players into three
categories – 1) Agent Beneficiaries, 2) Agent Contenders and 3) Agent Wildcards.

• Agent Beneficiaries are vendors where we have conviction that they can
either a) directly monetize AI Agents/Agentic architectures today or in the
near future such that it becomes a meaningful contributor to growth or b)
see their core business benefit materially from agentic adoption
• Agent Contenders are vendors that are in the right categories to see benefit
but require a high degree of execution and market maturity that is not yet
evident
• Wildcards are vendors where there is debate on whether agentic adoption
results in a tailwind to growth or proves disruptive to the core business

Morgan Stanley Research 5

M
Foundation

Based on our assessment of strategic positioning, product roadmaps and the ability
to execute to the opportunity, we highlight the following public companies as Agent
Beneficiaries.

AI infrastructure

• Microsoft (OW) – Covered by Keith Weiss

• Amazon (OW) – Covered by Brian Nowak
• Google (OW) – Covered by Brian Nowak

Cybersecurity

• CyberArk (EW) – Covered by Keith Weiss

• Okta (OW) – Covered by Keith Weiss
• Sailpoint (EW) – Covered by Keith Weiss
• Cloudflare (OW) – Covered by Keith Weiss

SaaS Platforms

• Salesforce (OW) – Covered by Keith Weiss

• HubSpot (OW) – Covered by Elizabeth Porter
• OneStream (OW) - Covered by Chris Quintero

Data Infrastructure

• Snowflake (EW) – Covered by Keith Weiss

• Palantir (EW) – Covered by Sanjit Singh

Workflow automation

• ServiceNow (EW) – Covered by Keith Weiss

• Atlassian (OW) – Covered by Keith Weiss

6
M
Foundation

Exhibit 5: Vendors Best Positioned to Benefit from Agentic AI Opportunity

Source: Morgan Stanley Research

Agent Pricing: Multiple Approaches But No Market Consensus Yet. AI Agent

pricing today is highly fragmented, as companies explore multiple approaches to
best capture their fair share of the value being created by their Agentic offerings.
Current approaches can be categorized along two dimensions – activity-based vs
outcome-based models, and fixed vs variable pricing structures.

In the activity-based segment, companies like Cognition Labs and Harvey AI have
adopted an "employee replacement" type model charging a fixed price per agent.
Microsoft, OpenAI, and Anthropic utilize a "consumption model" with variable
pricing based on agent actions/tokens. In the outcome-based segment, Salesforce
and Artisan implement a "process automation model" charging on a per completed
workflow basis (e.g., an Agentforce conversation regardless of successful outcome).
Zendesk and Sierra employ a "results-based model" where customers only pay if the
desired outcome was achieved. Meanwhile, vendors like Atlassian, SerivceNow and
HubSpot are taking a hybrid approach by embedding agent capabilities within higher
subscription tiers – with usage limits and ability to overconsume with additional pay.

Today most organizations still rely on traditional seat-based models, though

increasingly with usage-based components, while workflow and outcome-based
approaches continue to gain traction. This diversity in pricing strategies reflects the
nascent stage of the enterprise opportunity and underscores that the market is still
in a discovery phase in terms of settling on a pricing model for agentic solutions.

Morgan Stanley Research 7

M
Foundation

Exhibit 6: Most Companies Still Use Traditional Seat-Based Models, Though

Increasingly Combined with Usage-Based Component. Workflow and Outcome-
Based Models are Gaining Traction

Source: https://www.growthunhinged.com/p/ai-agent-pricing-framework, Morgan Stanley Research

The Obstacles to Widespread Adoption Are Significant. While most enterprises

today are looking for ways to leverage GenAI, widespread adoption of Agentic AI
across organizations will inevitably take time due to several unresolved obstacles.
First, the indeterminate nature of models today means that accuracy and
hallucinations remain an issue, making it difficult for businesses to fully trust AI-
driven decision making. Second, the lack of adequate security guardrails raises
serious concerns about the potential exfiltration of proprietary data or IP by bad
actors. Third, the absence of agreed-upon standards, protocols, and frameworks for
building multi-agent applications hinders interoperability and slows development.
Fourth, data quality issues and a fragmented data and application silos limit the
ability to effectively deal with complex business processes. Lastly, agentic systems
lack sufficient semantic understanding of proprietary business data, especially given
the unique processes and data nomenclature of each customer organization.
Collectively, these obstacles mean that enterprises will likely proceed in a deliberate
fashion as they work towards broader adoption of Agentic AI.

8
M
Foundation

Agents Knocking on the Door Presentation

Morgan Stanley Research 9

M
Foundation

Morgan Stanley Research

AI Agents Knocking at the Door

June 6, 2025

Sanjit Singh
[email protected]
(415) 576-2060

Keith Weiss
[email protected]
(212) 761-4149

Oscar Saavedra
[email protected]
(212) 761-0827

Morgan Stanley does and seeks to do business with companies covered in Morgan Stanley Research. As a result, investors should be aware that the firm may have a conflict of interest
that could affect the objectivity of Morgan Stanley Research. Investors should consider Morgan Stanley Research as only a single factor in making their investment decision.
For analyst certification and other important disclosures, refer to the Disclosure Section, located at the end of this report.

10
M
Foundation

MORGAN STANLEY RESEARCH

Executive Summary

Morgan Stanley Research 11

M
Foundation

MORGAN STANLEY RESEARCH

Executive Summary

• Bringing Agency to Software: Over the last several decades software has been developed based on the
What’s the Big presumption that humans play the central role in executing tasks, workflows and processes. The adoption of
Idea? AI Agents and Agentic Architectures means more and more of the software development effort will utilize a
framework where AI agents will be the entity executing tasks and making decisions.

• Bracing for Change: AI agents represent the next phase of a long-standing evolution in software towards
greater abstraction of the tools used to automate business and consumer processes. Like past waves, the
Implications transition towards AI Agents will result in greater automation, productivity and efficiency, but also force the
software industry to evolve business models and institute new approaches to pricing.

• A narrow view on the Agentic AI market points to a $6 billion opportunity today, expected to reach $20 billion
by 2028
Opportunity
• An expansive view on the Agentic AI market points to a $52 billion opportunity today, expected to reach $102
billion by 2028.

• Hyperscalers and AI infrastructure providers

• AI Model providers extending into the application layer
• Security vendors enabling authorization, threat detection, vulnerability/posture management and policy
Beneficiaries enforcement
• Data management providers
• Workflow and process automation providers
• SaaS Platforms

• Indeterminate Nature of the Models: Accuracy and hallucinations are still a problem
• Lack of adequate security guardrails to ensure proprietary data or IP is not exfiltrated by bad actors
Risks • Lack of agreement on standard, protocols and frameworks to build multi-agent applications
• Lack semantic understanding of enterprise’s business data from one organization to the next
• Limited reasoning capabilities to navigate complex business processes
3

12
M
Foundation

MORGAN STANLEY RESEARCH

What is an AI Agent?

What is an An AI Agent is a system (software program) that understands and interact with its environment (receive inputs), collect data, make
AI Agent decisions, executes tasks and meet specified goals without using pre-determined rules.

Autonomous AI Agents are systems that leverage large language models (LLMs) to plan and execute a task or process without a
Autonomous high degree of human interaction. These agents distinguish themselves from standard GenAI chatbots by combining tools (i.e.,
databases, websites, APIs, and even other agents) with memory of past interactions to chain multiple thoughts and actions
AI Agent together, enabling them to autonomously work toward defined objectives while drawing on both external information and learned
experiences to produce relevant outputs.

Agentic Architecture generally refers to the idea of bringing together multiple engines for understanding queries/requests,
Agentic evaluating strategies for solving problems, accessing necessary data and executing the planned actions. Under this architecture, a
large language model (LLM) typically serves as a reasoning engine, connected to tools and memory. Tools help connect the
Architecture LLM to other sources of data or computation and can be used to take actions (run code, modify files, etc.). Memory (short or long
term) helps the agent remember previous interactions with either humans, other agents, or tools.

4
Source: Company Data, Morgan Stanley Research, www.simform.com

Morgan Stanley Research 13

M
Foundation

MORGAN STANLEY RESEARCH

How Big is the Market Opportunity?

Narrow Evolution of Process Automation: A narrow view on the Agentic AI market points to a $6 billion
View opportunity today, expected to reach $20 billion by 2028

Expansive Broad Expansion of Application Functionality: An expansive view on the Agentic AI market points to a
View $52 billion opportunity today, expected to reach $102 billion by 2028.

• Increased demand from enterprises to further automate processes.

Key Drivers • Ongoing Improvements in foundational models results in more business value being realized, driving further demand.
• Continued declines in the price of intelligence which raises ROI for AI initiatives driving more demand for agents.

5
Source: IDC Worldwide Intelligent Process Automation Software Forecast

14
M
Foundation

MORGAN STANLEY RESEARCH

System of Record (SoR) vs System of Engagement (SoE) AI Agents

SoR vs. SoE High level framework for organizing the agent landscape is the distinction between System of
Ai Agents Record style agents from System of Engagement style agents.

System of Record (SoR) Agent:

• Automates transactions within one application or
domain
• For example, Saleforce’s Agentforce automated the
interactions of a Call Center agent

System of Engagement (SoE) Agent:

• Automates business processes spanning multiple
applications or systems
• For example, an HR Onboarding workflow spanning
Payroll, Benefits, IT Procurement systems to
complete a workflow

6
Source: Morgan Stanley Research

Morgan Stanley Research 15

M
Foundation

MORGAN STANLEY RESEARCH

System of Record (SoR) vs System of Engagement (SoE) AI Agents

SoR vs. SoE One framework for organizing the agent landscape is the distinction between System of Record
Ai Agents style agents from System of Engagement style agents.

SoR Agent SoE Agent

Role: Functional Supervisor

Automation Focus: Domain or Role Specific Workflow / Process-Oriented

Value Proposition: Labor Displacement VS System Utilization

More Direct - Automation of Indirect - Time Savings on

Quantification of ROI:
Role Broader Workflows
Difficulty of Harder - New Tech Plus Broader
Hard - New Tech Leaning Curve
Implementation: Integrations
Narrow - Defined by Role of Broad - Defined by Scope of
Market Opportunity:
Domain Workflows

7
Source: Morgan Stanley Research

16
M
Foundation

MORGAN STANLEY RESEARCH

The Emerging AI Agent Operating System

Key
The emerging AI Agent Operating System (OS) consists of: 1) the LLM, 2) Tools, 3) Access to
Elements of
Memory and 4) the ability to plan
the Agent OS

Memory
Long-term Memory
Short-term Memory

AI Agent
Tools Planning
Code Interpreter LLMs Chain of Thought
Calculator
Reflexion
Web Search
Self-critique
APIs
Open AI Llama Claude Deep Seek Subgoal Decomposition
Data Sources

Action

8
Source: Lilian Weng’s Lil’Log (https://lilianweng.github.io/)

Morgan Stanley Research 17

M
Foundation

MORGAN STANLEY RESEARCH

Current Market Landscape

Agentic AI market landscape extends from vendors enabling Agent development to vendors
Bottom Line
providing purpose-built and general-purpose Agents

9
Source: Morgan Stanley Research

18
M
Foundation

MORGAN STANLEY RESEARCH

Where in the Stack Will Value Accrue?

The most value likely accrues to the hyperscalers and AI infrastructure layer as AI Agent
Bottom Line deployments likely result in significant utilization of significant compute resources

• Other Parts of the Stack Likely to Benefit:

• Model Providers: Serving as the “brain” of any Hyper-
agentic solution gives license for model providers scalers /AI
to own more of the AI app stack Infra
• Security, Governance & Guardrails: The ability Model Providers
to infuse trust, reliability and accuracy while
enforcing policy present significant opportunities Security & Governance
for monetization
• Data Infrastructure: A necessary requirement to Data Infrastructure
build useful, effective and accurate agentic
systems require a modern data infra that make Workflow Automation
proprietary unstructured and structured data
easily governed and accessible End-to-End AI Lifecycle Platforms

• Workflow Automaters: Software providers that SaaS Platforms

are ingrained in customers’ key workflows and
business processes can create substantial value Monitoring & Observability
by further automating those workflows or re-
engineering altogether with agentic AI System of Record Providers

Tools

Agent Orchestration, Frameworks & Protocols

10
Source: Morgan Stanley Research

Morgan Stanley Research 19

M
Foundation

MORGAN STANLEY RESEARCH

Who is Poised to Capture the Agentic Opportunity?

Software vendors best positioned to benefit based on a) ability to monetize AI Agents/Agentic
Bottom Line
architecture or b) core business benefitting from enterprise adoption

• Agent Beneficiaries are vendors where we have

conviction that can either a) directly monetize AI
Agents/Agentic architectures today or in the near future
such that it to becomes a meaningful contributor to
growth or b) see their core business benefit materially
from agentic adoption
• Agent Contenders are vendors that are in the right
categories to see benefit but require a high degree of
execution and market maturity that is not yet evident
• Wildcards are vendors whose where there is debate on
whether agentic adoption results in a tailwind to growth or
proves disruptive to the core business

11
Source: Morgan Stanley Research

20
M
Foundation

MORGAN STANLEY RESEARCH

How Will AI Agents Be Priced?

Coming from an era of seat-based pricing, agentic pricing is taking on several forms with no clear
Bottom Line
consensus at this early stage

12
Source: https://www.growthunhinged.com/p/ai-agent-pricing-framework, Morgan Stanley Research

Morgan Stanley Research 21

M
Foundation

MORGAN STANLEY RESEARCH

Part I
Key Debates

13

22
M
Foundation

MORGAN STANLEY RESEARCH

Debate #1: How Big is the Opportunity?

A conservative view on the Agentic AI market points to a $6 billion opportunity today, expected to
Bottom Line
reach $20 billion by 2028

• IDC definition of AI-powered automation includes agentic automation, intelligent document processing (IDO), business process-
specific specialized models, and predictive analytics

14
Source: IDC Worldwide Intelligent Process Automation Software Forecast

Morgan Stanley Research 23

M
Foundation

MORGAN STANLEY RESEARCH

Debate #1: How Big is the Opportunity?

An expansive view on the Agentic AI market points to a $52 billion opportunity today, expected to
Bottom Line
reach $102 billion by 2028

• System of Record (SoR) AI Agent: Aligning the SoR

market opportunity with IDC’s labor-centric automation
market segment
• $16B market opportunity today → $33B by 2028
• Labor-centric automation includes business
process automation and robotic process
automation
• Functionalities typically support people-oriented
activities that require manual steps or human
decision-making (i.e., loan approvals, new
employee onboarding)
• System of Engagement (SoE) AI Agent: Aligning the
SoE market opportunity with IDC’s system-centric,
decision-centric, and business value engineering market
segments
• $36B market opportunity today → $69B by 2028
• Functions typically involve back-end automation
of two or more systems to support interoperability,
exchange info, and process transactions, along
with rules-based decision automation

While our definitions of SoR/SoE AI Agents do not exactly

match IDC’s market segment definition, we believe these
serve as close proxies
15
Source: IDC Worldwide Intelligent Process Automation Software Forecast

24
M
Foundation

MORGAN STANLEY RESEARCH

Debate #1: How Big is the Opportunity?

As Agentic AI solutions mature, more of the Application Software market should open up as
Bottom Line
potential opportunity

16
Source: IDC Software Tracker Forecast

Morgan Stanley Research 25

M
Foundation

MORGAN STANLEY RESEARCH

Debate #2: Where Will Value Accrue?

The most value likely accrues to the hyperscalers and AI infrastructure layer as AI Agent
Bottom Line deployments likely result in significant utilization of significant compute resources

• Other Parts of the Stack Likely to Benefit:

• Model Providers: Serving as the “brain” of any Hyper-
agentic solution gives license for model providers scalers /AI
to own more of the AI app stack Infra
• Security, Governance & Guardrails: The ability Model Providers
to infuse trust, reliability and accuracy while
enforcing policy present significant opportunities Security & Governance
for monetization
• Data Infrastructure: A necessary requirement to Data Infrastructure
build useful, effective and accurate agentic
systems require a modern data infra that make Workflow Automation
proprietary unstructured and structured data
easily governed and accessible End-to-End AI Lifecycle Platforms

• Workflow Automaters: Software providers that SaaS Platforms

are ingrained in customers’ key workflows and
business processes can create substantial value Monitoring & Observability
by further automating those workflows or re-
engineering altogether with agentic AI System of Record Providers

Tools

Agent Orchestration, Frameworks & Protocols

17
Source: Morgan Stanley Research

26
M
Foundation

MORGAN STANLEY RESEARCH

Debate #2: Assessing Where Value Accrues in the Stack

Layer Definition Rationale Examples Monetization

AI Infrastructure providers have a significant opportunity in providing the compute to
Amazon AWS,
Includes the compute, network, storage and enable the real-time inference that deep reasoning agents require. This gives Google Cloud,
Hypescaler / AI Infra model serving capabilites for the AI models that leaders in this category license to not only provide the infrastructure but also the
Microsoft Azure, Very High
power agentic architecture data retrieval, security, governance and policy enforcement capabilities required to
Coreweave,
operate,manage and secure agentic applications. Lambda
AI model providers extend the frontier around reasoning and intelligence that will
Open AI,
Includes the state-of-the-art language, image,
make AI agents more capable over time. This positioning provides the opportunity
Anthropic, Llama,
AI Model Providers video and audio AI models both commercial and Very High
to extend into areas such as agentic frameworks, data processing and integration
Deep Seek,
open source
so as to evolve into modern application platforms. Mistral, Grok
Palo Alto
We see a significant monetization opportunity in making agentic applications more
Includes the state-of-the-art language, image, Networks,
trustworthy, reliable, accurate and secure. Agent architectures create a large
Security & Governance video and audio AI models both commercial and Crowdstrike, Okta, Very High
surface area for cyber attacks including data exfiltration, supply chain attacks,
open source Sailpoint,
prompt injection and open source contamination.
CyberArk, Rubrik
For AI agents to be useful, they need governed access to the most important data
Includes the data management software that sources in real-time. We see two key opportunities for data infrastructure Oracle, Snowflake,
Data Infrastructure enables and govern access to an organization's providers. First, in helping make agentic architectures possible by modernizing a MongoDB, High
structured, semi-structured unstructured data customer's data estate and improving data quality. Second, by enabling fast, Confluent
reliable and accurate information retrieval at runtime.
With AI agents embedded in workflow automation platforms, the unstructured
Workflow automation reduces manual work by
components of common business processes can be more fully automated. As ServiceNow,
automating the flow of tasks, documents and
Workflow workflow automation platforms span across key enterprise systems, they become Atlassian, Monday High
data involved in common work related activities
well-positioned to serve as an orchestration/management layer for both 1st and 3rd Appian, UiPath
and business processes
party agents.
Agents within AI lifecycle platforms automate the workflows and tasks associated
Platforms that automate the lifecycle for building
with developing and deploying AI applications including building data pipelines,
+ running AI applications including data
AI Lifecycle Platforms integrating enterprise data with LLMs, building and evaluating models and Palantir High
integration, data processing, ontology mapping,
visualizing data. Furthermore, agents can be embedded into external applications
model building/evaluation and app deployment
to enhance customer engagement and to enable highly interactive experiences.

18
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 27

M
Foundation

MORGAN STANLEY RESEARCH

Debate #2: Assessing Where Value Accrues in the Stack

Layer Definition Rationale Examples Monetization

With agents, SaaS platforms will be able drive much deeper levels of automation
Includes incumbent SaaS platforms who serve
and efficiency for each business function or domain they operate. AI Agents add
as system of record across multiple domains Salesforce, Adobe,
SaaS Platforms value by unlocking the full capabilities of SaaS platforms. However, the challenge Medium
resulting in these platforms operating multiple HubSpot
will be pricing as SaaS providers have often priced per seat or per user heading
data silos for their end customers
into an era where seats (at least for some functions) may stop growing or decline
The adoption of agentic architectures increases the complexity of the app
software that monitors the health and environment and expands the surface area for monitoring. Furthermore,
performance of the digital business including the
observability agents can automate the investigation process and kick off Datadog,
Observability Medium
underlying infrastructure, applications and endremediation workflows typically executed by engineering staff. The debate will be Dynatrace
user activity whether incumbents capture the opportunity or whether there will be a need for a
new breed of AI-native observability solutions.
AI agents, simplify access to systems of record and make it easier for customers
Systems of record software are information
to build operational and analytical applications on top of these core business
retrieval systems that serve as the source of
systems. In addition, agents can replace human labor for common tasks such as Workday, ADP,
System of Record truth and central repository for data of specific Medium
cash collection associated with key business functions. However, to fully automate Intuit
domain ensuring consistency and accuracy
business processes that span multiple systems, requires agents that can
across the entire organization
coordinate across multiple systems
Tools make Ai agents useful by giving Agents the ability to take action in the face of Calculators, web
Encompass the set of actions that an AI agent
query or trying to solve a problem or task. Tools can include calendars, calculators, search, code
Tools can take by calling specific services or functions Low
code interpreters, web search, enterprise search, APIs, databases and much interpreters,
it has access to.
more. databases
The abstraction and standardization benefits that application frameworks and
Includes frameworks and protocols to build multi-
management/orchestration tools provide play a crucial role in accelerating the pace
agent applications, foster agent to agent
of development of modern applications. However, the monetization in this area of Autogen, MCP,
Frameworks & Protocolscommunication, enable multi-agent Low
software has been mixed looking back over the past few decades, particularly as A2A, ACP
orchestration and connect LLMs to external tools
open source technologies have played a more dominant role within application
and data sources
infrastructure

19
Source: Company Data, Morgan Stanley Research

28
M
Foundation

MORGAN STANLEY RESEARCH

Debate #3: Which Public Companies Are Positioned to Capture the Opportunity?
Software vendors best positioned to benefit based on a) ability to monetize AI Agents/Agentic
Bottom Line
architecture or b) core business benefitting from enterprise adoption

• Agent Beneficiaries are vendors where we have

conviction that can either a) directly monetize AI
Agents/Agentic architectures today or in the near future
such that it to becomes a meaningful contributor to
growth or b) see their core business benefit materially
from agentic adoption
• Agent Contenders are vendors that are in the right
categories to see benefit but require a high degree of
execution and market maturity that is not yet evident
• Wildcards are vendors whose where there is debate on
whether agentic adoption results in a tailwind to growth or
proves disruptive to the core business

20
Source: Morgan Stanley Research

Morgan Stanley Research 29

M
Foundation

MORGAN STANLEY RESEARCH

Debate #3: Which Vendors Are Best Positioned to Win?

Agent Beneficiaries – Hyperscalers

• In our 1Q25 CIO Survey, CIO’s indicated Microsoft as the top vendor expected to be used for Agentic AI strategies
• Microsoft’s focus and expertise on security and data privacy create critical trust within vast existing customer installed
base to develop and deploy agents responsibly
• Microsoft’s ubiquitous productivity & business apps allow for deep integrations w/ Enterprise workflows and data; its
Microsoft strong developer ecosystem and platform offerings provide a natural adoption path of Microsoft’s Agentic AI products
• Microsoft benefits from its comprehensive AI portfolio and its breadth of Agent specific offerings, including: GitHub
Copilot’s asynchronous coding agent; Azure AI Foundry Agent Service, Microsoft 365 Copilot Studio / Agent Build,
Agents for Microsoft Teams, M365 Agents toolkit for Visual Studio, M365 Agents SDK, Data Agents in Microsoft Fabric,
SRE agents (among others)

• AMZN continues to be the share leader within the public cloud market, with the largest base of customers to cross sell
GenAI capabilities into. AWS has already generated multiple billions of dollars in AI revenue.

Amazon • In our 1Q25 CIO Survey, Amazon ranked as the second largest gainer of incremental share of both overall IT budgets
and GenAI workloads.
AWS
• AMZN offers Amazon Q to businesses, integrating with a user’s AWS account, business intelligence tools, contact
center and business apps to provide comprehensive and personalized answers to user questions as well as perform
tasks on behalf of the user.

• In our 1Q25 CIO Survey, Google Cloud was the third largest gainer of incremental share of IT budgets as enterprises
shift to the cloud in the next three years, trailing Microsoft and Amazon

Google • Google has announced various agentic offerings and shipping products at an accelerated pace, including Project
Mariner (which is coming to Vertex AI), Gemini Code Assist, Vertex AI Agent Builder, and Gemini for Google
Cloud Workspace
• Gemini 2.5 Pro is currently the leading model across a variety of benchmarks and Google’s integration of Gemini
across GCP could improve capabilities and enterprise adoption
21
Source: Company Data, Morgan Stanley Research

30
M
Foundation

MORGAN STANLEY RESEARCH

Debate #3: Which Vendors Are Best Positioned to Win?

Agent Beneficiaries – Model Providers

• OpenAI has established itself as a frontrunner in Agentic AI development, with several foundational models serving as
the intelligence backbone for agentic systems. The company has released a complete development ecosystem for
building agentic applications including 1) Agents SDK, 2) Responses API, and 3) built-in tools integration
• The company has moved beyond theoretical frameworks and delivered practical agentic applications that demonstrate
its potential including 1) Operator, an autonomous AI agent capable of performing complex tasks through web browser
OpenAI interactions, 2) Codex, an agentic web-based coding tool, and 3) Deep Research, an agent that conducts in-depth web
searches and analyses to generate comprehensive reports
• OpenAI has significant financial backing enabling it to innovate, having raised a total of $48 billion in private funding
with the latest record-setting $40 billion funding round at a $300 billion valuation – would make it the fourth largest
publicly traded software company after MSFT ($3.4 trillion), ORCL ($560 billion), and PLTR ($318 billion)

• Anthropic has emerged as a formidable contender to OpenAI, with differentiated technological capabilities, strategic
partnerships, and a governance framework uniquely suited for enterprise adoption
• The company’s Claude models provide the cognitive backbone for advanced agentic systems, including Claude 3.7
Sonnet which introduced a breakthrough “hybrid reasoning” framework that enables users to control the computational
Anthropic resources used, optimizing performance based on task complexity. The model family (Haiku, Sonnet, Opus) offers
right-sized solutions across use cases, from real-time customer service agents to longer term tasks with many steps
• Strategic partnerships – AWS and Accenture – and constitutional AI framework is helping drive enterprise adoption,
particularly in highly regulated industries. Partnership with AWS and Palantir to provide U.S. intelligence and defense
agencies access to Claude, validating security credentials for sensitive applications

22
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 31

M
Foundation

MORGAN STANLEY RESEARCH

Debate #3: Which Vendors Are Best Positioned to Win?

Agent Beneficiaries – Cybersecurity

• CyberArk has meaningfully invested in expanding its machine identity and agentic security capabilities through its
acquisition of Venafi and the announced launch of its Secure AI Agents solution (see here for additional details).

CyberArk • Given the potential scalability and broader AI agent market opportunity, there’s likely room for multiple winners within
the Identity Security landscape, though CyberArk’s foundation in PAM positions it well with the IT admins that will
ultimately be managing and responsible for the expansion of AI agents across enterprise environments while the
company’s entrance into machine identity increases platform capabilities to better address non-human identities.

• IGA solutions are focused on helping organizations set and enforce policies around who has access to what
applications and data in an organization and as AI agents are deployed at scale across an enterprise, they will need an
increasing level of access to both data and applications to complete more complex tasks. With SailPoint’s foundation in
SailPoint IGA, existing human identity security capabilities should translate well to non-human’s
• Announced its “Agent Identity Security” solution in March with capabilities including 1) managing lifecycle of AI agents
with automated governance, and 2) enforcing access certifications to prevent unauthorized access and security risks.
• A leader in human authentication, a growing presence within governance, and the company’s “Auth for GenAI” solution
highlight the continued expansion of Okta’s capabilities across multiple identity security growth vectors.
• While Okta is taking a different approach to positioning itself within the agentic marketplace (relative to CyberArk and
Okta SailPoint), its leadership in authentication and success with newer product cycles highlight the company as another
potential winner as the use of AI agents across the enterprise continue to expand.
• Okta’s “Auth for GenAI” solution includes 1) user authentication, 2) calling APIs on the user’s behalf, 3) async
authentication, and 4) fine grained authorization for retrieval augmented generation (RAG).
• In early innings of monetizing its multi-product platform and has emerged as a potential GenAI winner as orgs. look to
adopt and experiment with LLMs. By leveraging its “Edge Network” with 330+ PoPs globally, it’s well-positioned to gain
share given increased AI inference at the edge, especially as it relates to cost, latency and data security/residency.
Cloudflare • Cloudflare’s Developer Services Platform already has >3 million developers and continues to see growing adoption
across R2 Object Storage, Workers and AI gateway for training, inference and monitoring AI workloads.
• Cloudflare’s Workers AI, a serverless GPU-powered platform on Cloudflare's global network, provides all the
foundational elements of running AI inference on GPUs in 190+ cities. 23
Source: Company Data, Morgan Stanley Research

32
M
Foundation

MORGAN STANLEY RESEARCH

Debate #3: Which Vendors Are Best Positioned to Win?

Agent Beneficiaries – Data Infrastructure

• Snowflake is poised to benefit from the deployment of AI agents and agentic architectures in two fundamental ways.
First, as one of the popular stores of business data and increasingly semi-structured and unstructured data, Snowflake
will be a key source of context and information for a variety of 3rd party and custom developed AI agents. Second,
Snowflake is also well positioned to monetize agentic architectures with its Cortex Agents.
• Cortex agent leverage two key AI “tools” from Snowflake – 1) Cortex Analyst and 2) Cortex Search. Cortex agents call
Snowflake Cortex Analyst in scenarios when the agent needs to respond to queries requiring structured data leveraging Cortex
Analyst’s text-to-sql capabilities. In cases, where the agent needs to respond to an unstructured data request (such as
document or an invoice) Cortext Search is called given its hybrid search (semantic + keyword) capabilities.
• Snowflake’s agents can be embedded in any application and the core value proposition is the ability to respond to
structured and unstructured queries from a single platform.

• As part of its Foundry platform for commercial customers, Palantir AIP allows customers to securely deploy LLMs
within organizations and provides a set of integrated capabilities including data integration, ontology mapping,
management services, workflow automation, application deployment and many others to build AI applications.
• AIP supports the building and deployment of AI agents that can then be embedded within AIP applications via the AIP
Palantir Agent Studio module within AIP
• With Agent Studio, customer can build agents of varying complexity including simple ad-hoc agents, task-specific
agents, agentic applications (that can manage and update state) and fully automated agents that can be delegated
tasks to fully execute complex workflows.

24
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 33

M
Foundation

MORGAN STANLEY RESEARCH

Debate #3: Which Vendors Are Best Positioned to Win?

Agent Beneficiaries – Workflow Automation

• ServiceNow’s expansion in the scope of workflows automated and ability to combine AI, data, and workflow automation
position it to become the de facto “AI operating system for enterprises”.
• AI agents and their ability to solve more complex use cases will be based upon the data they’re built upon and
ServiceNow’s expansive workflow automation capabilities (IT, HR, customer service, etc.) combined with its CMDB
ServiceNow serving as the system of record for enterprises’ IT infrastructure (>85% of Fortune 500 are customers) provide the
company with a robust data asset that it can leverage to offer differentiated AI agents.
• Building upon this data advantage, ServiceNow’s Workflow Data Fabric offering provides customers with the ability to
connect their data residing outside of ServiceNow into the platform, unifying customer data and expanding the number
of use cases AI agents can address.

• Atlassian’s core products are deeply embedded in the daily workflows of enterprises and teams, giving the company a
unique vantage point on how modern organizations collaborate, manage projects, and share knowledge. This
foundation, along with a highly efficient innovation engine, positions it well to integrate agentic AI across core
workflows and benefit from the emerging opportunity
Atlassian • Atlassian has embedded Rovo and Atlassian Intelligence – its AI/Agentic AI functionalities – into its core products (Jira,
Confluence, and Jira Service Management). We view this approach as the right strategic move for driving near-term
adoption and usage to secure a better positioning for the long-term agentic AI opportunity
• Today there are >1.5 million MAUs on Rovo and Atlassian Intelligence, up from >1 million a quarter ago

25
Source: Company Data, Morgan Stanley Research

34
M
Foundation

MORGAN STANLEY RESEARCH

Debate #3: Which Vendors Are Best Positioned to Win?

Agent Beneficiaries – SaaS Platforms

• Salesforce remains a key play on the incumbency advantage to operationalizing AI agents, with competitive
differentiation across: 1) Defined Role, where Salesforce can extrapolate clearly outlined jobs for the agents that exist
inside an organization, de-risking the process & time to defining the experience; 2) Data, where Data Cloud ingests
structured, unstructured, metadata, & federated data both internally from Salesforce and from external sources
(Snowflake, Databricks, etc), to drive knowledge behind the agents; 3) Actions, or the workflows needed to perform a
Salesforce job, where Salesforce can capitalize on years of experience building a broad swath of workflows, all while leveraging
the Trust Layer to place ethical and accuracy guardrails around each step of the agent process
• Salesforce’s solid Agentic positioning is growing increasingly apparent with building Agentforce momentum: as of
F1Q26, the company sustained 120% YoY growth in Data + AI ARR (~$100M of which is specific to Agentforce) &
reported ~8k Agentforce deals, up from ~5k in 4Q25
• HubSpot is well positioned to address the agent opportunity given: 1) deep data sets combining rich structured and
unstructured data; 2) a unified platform approach that provides context across data, engagement (myriad different
hubs), and action (AI-powered execution); and 3) a vibrant agent ecosystem developed through the company’s
Agent.ai network. Further, HubSpot’s value proposition as an easy-to-use, out of the box solution with AI embedded
across the platform allows for quick time-to-deployment and potential to scale over time
HubSpot • Strategically, HubSpot remains relentlessly focused on driving strong adoption and utilization with new products before
introducing direct monetization, a framework we believe should foster strong and durable monetization tailwinds ahead
• With usage and adoption statistics starting to materially inflect, HubSpot’s move to integrate credit-based pricing
strategies for the Customer Agent product bridges an intuitive hybrid Subscription/Consumption model which should
well align price to value for the solution
• AI Agents have a higher likelihood of initial adoption and are monetizable within domains that are: 1) Primarily cost
centers, where organizations can realize tangible ROI from reduced headcount; 2) Experiencing talent shortages,
where organizations are increasingly looking to technology as the solution for the labor gap; and 3) High levels of
manual data processing. Furthermore, given enterprise CFO buyers are inherently more risk-averse, we believe this
OneStream will allow for existing vendors in this space to benefit versus new vendors.
• OneStream has been developing AI/ML tools since 2017 (SensibleML) and recently launched multiple agents (Finance
Analyst, Operations Analyst, Search Analyst, Deep Analysis) that have seen strong initial customer feedback as the
agents can handle a lot of the manual and data intensive parts of their jobs.
26
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 35

M
Foundation

MORGAN STANLEY RESEARCH

Debate #3: Which Vendors Are Best Positioned to Win?

Gartner defines this segment as “tools that augment and extend Magic Quadrant for AI Applications in IT Service Management
IT Service Management workflows using AI.”
• They analyze ITSM data and metadata to provide advice and
actions on IT service desk and support activities.
• They can either be stand-alone solutions, capabilities within
an ITSM platform, or an add-on to an ITSM platform.
While this definition and solutions are not strictly “Agentic”, this
gives us a rough sense of which vendors are best positioned in
the ITSM space as this technology progresses and garners
adoption.
• ServiceNow offers “thousands” of pre-built AI agents for
“every” workflow and an AI Agent Studio to build fully
customized agents. The company also recently introduced its
AI Agent Orchestrator, a “control tower” that conducts
planning and leads collaboration between ServiceNow and
third-party AI Agents.
• Aisera’s Agent Assist is an Agentic AI-driven tool designed to
boost agent productivity by providing real-time, context-aware
answers and case summaries, recommend next best actions,
and draft knowledge-base articles. Embeds directly within
SaaS apps like ServiceNow, Salesforce, Zendesk, Jira,
Freshworks, etc.
• Moveworks’ platform, powered by its Agentic Automation
Engine, enables users to create AI Agents for automating
workflows across the organization.

27
Source: Gartner Magic Quadrant for AI Applications in IT Service Management

36
M
Foundation

MORGAN STANLEY RESEARCH

Debate #3: Which Vendors Are Best Positioned to Win?

Gartner defines this segment as “tools that assist in generating

and analyzing software code and configuration.” Magic Quadrant for AI Code Assistant

• They use LLMs fine-tuned for code and/or program-

understanding technology, and integrate into developer tools
like code editors, command-line terminals and chat interfaces.
• Developers can prompt it to generate, analyze, debug, fix,
and refactor code, to create documentation, and to translate
code between languages.
While this definition and solutions are not strictly “Agentic”, this
gives us a rough sense of which vendors are best positioned in
the Software Developer space as this technology progresses
and garners adoption.
• GitLab Duo provides AI capabilities across the software
delivery lifecycle, including 1) code generation, completion,
explanation, refactoring, 2) security vulnerability and root
cause analysis, and 3) value stream analytics. In addition,
GitLab is working on GitLab Duo Workflows, moving beyond
code assistant and into Agentic AI. It’ll behave like an
autonomous always-on agent monitoring projects to identify
and resolve
• GitHub Copilot developed by GitHub in collaboration with
OpenAI, provides real-time code suggestion, completions,
and assistance throughout the software delivery lifecycle. It
was first to market (ahead of GitLab) and is the most widely
used code assistant today.
• Google’s Gemini Code Assist
• Amazon CodeWhisperer
28
Source: Gartner Magic Quadrant for AI Code Assistant

Morgan Stanley Research 37

M
Foundation

MORGAN STANLEY RESEARCH

Debate #3: Which Vendors Are Best Positioned to Win?

Magic Quadrant for Enterprise Conversational AI Platforms

Gartner defines this segment as “software platforms used to

build, orchestrate and maintain multiple use cases and
modalities of conversational automation.”
• Most notably addresses customer service and support, IT
service desk, HR, and call center automation.
While this definition and solutions are not strictly “Agentic”, this
gives us a rough sense of which vendors are best positioned in
the conversational agent space (i.e., customer service, sales and
marketing, etc.) as this technology progresses and garners
adoption.
• Kore.ai offer an Agent Platform with pre-built and custom AI
Agent for internal support (i.e., HR, IT), external (i.e.,
customer service), and process automation.
• Cognigy offers customer service specific AI agents that can
be easily deployed in contact centers. It offers voice and chat
agents, as well as agent copilots for human service agents.

29
Source: Gartner Magic Quadrant for Enterprise Conversational AI Platforms

38
M
Foundation

MORGAN STANLEY RESEARCH

Debate #4: How Will Agents be Priced?

Many vendors focused on driving adoption and usage rather than generating meaningful revenue
Bottom Line from their AI Agents. Despite having a pricing model in place, most are experimenting with
different forms, including usage-based, workflow-based, and outcome-based.

• Fixed Monthly Fee: Customers pay a recurring fee for access to the AI agent, regardless of usage (i.e., Devin)
Per Agent • Tiered Pricing: Different pricing tiers offer varying levels of access and features
• Per-User Pricing: Customers pay a fee per user who has access to the AI agent (i.e., NOW)

• API Calls: Vendors charge based on the number of requests made to the AI agent’s API (CRM’s Flex Credits)
Per Action /
• Compute Time/Data Processed: Pricing is tied to the amount of processing power or data the AI agent uses
Consumption • Token-Based: Based on the number of tokens used for input and output by the AI agent (i.e., Anthropic, OpenAI)

Per Workflow • Per Conversation: Customers pay a fee for each AI agent interaction or conversation (i.e., CRM)

• Latest evolution of pricing, whereby customers pay based on the successful outcomes achieved by the AI
agent, such as resolved conversations, saved cancellations, or increased sales.
Per Outcome
• While not publicly available, Sierra emphasizes an “outcome-based pricing” approach for their customer service AI
agent.

• Access to AI Agents/features included in the higher subscription tier.

Embedded
• Current state of many vendors who are focusing on driving customer adoption and usage (i.e., TEAM and HUBS).
30
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 39

M
Foundation

MORGAN STANLEY RESEARCH

Debate #4: How Will Agents be Priced?

Most companies still use the traditional seat-based models, though increasingly combined with a
Bottom Line
usage-based component. Workflow and outcome-based models are gaining traction

31
Source: https://www.growthunhinged.com/p/ai-agent-pricing-framework, Morgan Stanley Research

40
M
Foundation

MORGAN STANLEY RESEARCH

Debate #4: How Will Agents be Priced?

Case Study: Workflow-based Pricing – Salesforce (Agentforce)

Initially launched with straightforward $2 per conversation model

• Conversation types include:
• ASA Messaging: Conversation window starts when the user
first sends a message and ends 24 hours later
• Sales Coach: Conversation is consumed when a user clicks
“Get Feedback” following a role play or a stand-and-deliver
coaching session
• Sales Development Rep (SDR): Conversation is consumed
when an SDR agent sends out the initial email to a lead.
Includes all subsequent activities carried out by the agent for
that lead
• Each conversation includes 200 Einstein Requests, with additional
requests drawing from a shared pool or incurring overage charges

Introduced three additional pricing models on May 15, 2025, including:

• Flex Credits: Offers granular consumption-based pricing at $0.10
per action (20 Flex Credits), allowing customers to pay only for
specific agent activities. Credits are sold in packs of 100K for $500
• Flex Agreement: Convert user licenses into Flex Credits or vice-
versa, allowing you to move your budget between people and AI
agents as needed
• Agentforce User Licenses, and add-ons with included Agentforce
usage: Unlimited employee-facing agent usage through traditional
per-user, per-month pricing. Pricing coming Summer 2025
32
Source: Company website

Morgan Stanley Research 41

M
Foundation

MORGAN STANLEY RESEARCH

Debate #4: How Will Agents be Priced?

Case Study: Outcome-based Pricing – Zendesk (AI Agents)

Pricing strategy exemplifies the evolving landscape of enterprise AI Traditional Subscription

monetization, blending traditional subscription with outcome-based
pricing
• Existing subscription plans embed AI capabilities, with
incremental features on higher tiers and add-ons
• Outcome-based pricing: Introduced in Sept 2024 as a model
directly dependent on the AI agent performance
• Pay $1.50-$2 per successfully resolved AI interaction –
with volume discounts available
• Exempts escalations to human agents from charges,
directly tying costs to business results
Outcome-based model addresses key customer pain points:
• Cost predictability: Avoid overpaying for underperforming AI
• Scalability: Unlimited AI capacity during peak periods w/o fixed Outcome-based Pricing
staffing costs
• ROI transparency: Compare cost per AI-resolved ticket to
human cost
We note some customer confusion over what constitutes an
“automated resolution”, with abandoned chats sometimes triggering
charges

33
Source: Company website

42
M
Foundation

MORGAN STANLEY RESEARCH

Debate #4: How Will Agents be Priced?

Case Study: Combined Model – HubSpot (Breeze)

Two-tiered model combining traditional subscription with

consumption-based credits

Understanding the Philosophy

• Given HubSpot’s strategic approach to driving healthy and consistent
adoption/utilization on new products (including Agents) before moving
to direct monetization, the company initially rolled out Breeze Agents in
Beta across all Platforms at September 2024’s INBOUND Conference

Direct Monetization Phase

• HubSpot Credits: As of June 2025, HubSpot is rolling out HubSpot
Credits, a pay-as-you-scale model that allows customers to pay for
usage-based features. Credits are utilized when an action is performed
that consumes them (ex, Service tickets resolved)

Key Considerations
• How Many Credits are Allocated?*
• Free: N/A
• Starter: N/A
• Pro: 3k credits/month
• Enterprise: 5k credits/month
• What Subscriptions/Seats are Eligible?
• All Hubs (Marketing, Sales, Service, Operations, Content)
across all tier levels (Starter, Pro, Enterprise)
• Only Paid & Partner seats (not free/view-only users)
• How do Users Purchase Incremental Credits Above What is Allocated?
• Small: 5k incremental credits/month, priced at $45/month
• Medium: 30k incremental credits/month, priced at $270/month
• Large: 100k incremental credits/month, priced at $900/month

34
Source: Company website

Morgan Stanley Research 43

M
Foundation

MORGAN STANLEY RESEARCH

Part II
How Did We Get Here?

35

44
M
Foundation

MORGAN STANLEY RESEARCH

What is an AI Agent?

What is an An AI Agent is a system (software program) that understands and interact with its environment (receive inputs), collect data, make
AI Agent decisions, executes tasks and meet specified goals without using pre-determined rules.

Autonomous AI Agents are systems that leverage large language models (LLMs) to plan and execute a task or process without a
Autonomous high degree of human interaction. These agents distinguish themselves from standard GenAI chatbots by combining tools (i.e.,
databases, websites, APIs, and even other agents) with memory of past interactions to chain multiple thoughts and actions
AI Agent together, enabling them to autonomously work toward defined objectives while drawing on both external information and learned
experiences to produce relevant outputs.

Agentic Architecture generally refers to the idea of bringing together multiple engines for understanding queries/requests,
Agentic evaluating strategies for solving problems, accessing necessary data and executing the planned actions. Under this architecture, a
large language model (LLM) typically serves as a reasoning engine, connected to tools and memory. Tools help connect the
Architecture LLM to other sources of data or computation and can be used to take actions (run code, modify files, etc.). Memory (short or long
term) helps the agent remember previous interactions with either humans, other agents, or tools.

36
Source: Company Data, Morgan Stanley Research, www.simform.com

Morgan Stanley Research 45

M
Foundation

MORGAN STANLEY RESEARCH

Evolution Over Time

The concept of “rule-based systems” in AI can be traced back to the 1970s, when the earliest “expert systems” were built on logical
Rule-Based rules inspired by how experts in various fields (i.e., medicine, law) used their knowledge to make decisions.
Systems The most famous example (MYCIN) developed at Stanford was designed to diagnose bacterial infections and recommend treatments
based on a set of predefined rules.

Natural During the late 1980s, natural language processing (NLP) – a subdivision of AI which makes human language understandable to
computers – experienced an evolutionary leap as a result of steady increase in computational power and the use of new machine
Language learning algorithms – which focused primarily on statistical models as opposed to models like decision trees. During the 1990s, statistical
Processing models for NLP rose dramatically.

Intelligent In the early 1990s, AI research shifted its focus to “intelligent agents” – sometimes called agents or bots – which could be used for news
retrieval services, online shopping, and web browsing. This era saw the rise of machine learning (data-driven approaches and neural
Agents networks) and emergence of intelligent agents designed to operate autonomously.

In the 2000s, with the use of Big Data programs (i.e., Apache Hadoop, NoSQL databases) along with advances in machine learning and
Machine data mining to extract information from complex data sets, agents gradually evolved into digital virtual assistants and chatbots. These
digital virtual assistants started as convenient sources of information about the weather, the latest news, and traffic reports. Chatbots,
Learning although similar to virtual digital assistants in many ways, could talk to real people and were often used for marketing, sales, and
customer service.

Advances in NLP, machine learning, decision-making capabilities and access to massive amounts of data have turned digital virtual
assistants and chatbots into truly useful tools. Current AI Agents combine multiple approaches and capabilities, including 1) large
Modern language models (LLMs) providing natural language understanding and generation; 2) multi-modal processing allowing agents to work
AI Agents with text, images, audio, and other data types; 3) reasoning and planning capabilities through integration of LLMs and techniques like
chain-of-thought prompting; 4) integration with external tools and APIs to perform tasks; and 5) use of shorter-term and long-term
memory for continuity, self-reflection, and performance improvement over time.

37
Source: Morgan Stanley Research

46
M
Foundation

MORGAN STANLEY RESEARCH

Key Milestones and Recent Breakthroughs

38
Source: Morgan Stanley Research

Morgan Stanley Research 47

M
Foundation

MORGAN STANLEY RESEARCH

RPA vs AI Agents?

Key Robotic Process Automation uses pre-defined rules to automate tasks that involve structured
data. In contrast AI agents are able to automate tasks and processes using indeterministic,
Distinction reasoning and context while showing the ability to adapt to change

Robotic Process Automation (RPA) VS AI Agents

AI Agents are autonomous systems that perceive their

RPA uses software robots or “bots” to automate repetitive, environment, analyze data, and take actions to achieve
rule-based tasks by mimicking human interactions with digital specific goals. They use machine learning, natural language
systems. It operates based on predefined workflows and processing, and reasoning to adapt, learn from experiences,
structured data. and make decisions without explicit human intervention.

• Decision Making: Follows pre-defined rules and decision • Decision Making: Autonomous, undeterministic based on
trees reasoning and context
• Data Handling: Structured, predefined inputs • Data Handling: Structured and unstructured, dynamic
• Orchestration: Centralized control, follows fixed inputs
workflows • Orchestration: Decentralized, can self-organize and
• Interaction: UI-based, mimics human clicks and collaborate
keystrokes • Interaction: Conversational, API-driven, and reasoning-
• Fault Tolerance: Low; fails when encountering based
unexpected inputs • Fault Tolerance: High; can adjust to errors and
• Learning: No learning; runs static rules to execute routine uncertainties
tasks • Learning: Continuous learning allows for reasoning,
problem-solving and decision making

39
Source: Morgan Stanley Research

48
M
Foundation

MORGAN STANLEY RESEARCH

Levels of Agentic Automation

No Agentic behavior – pure rule-based automation. Tasks follow a fixed, deterministic process with no planning or
Level 0 execution control. Examples include Data entry and web scrapping. i.e., RPA

Basic AI-enhanced automation where LLMs assist in decision-making but within predefined steps. Limited benefits
Level 1 over traditional automation. Examples include routing customer support emails.

Task-specific AI assistant capable of using tools, interpret intent, execute actions like summarization or content
Level 2 generation, but follow static, short-term plans. Examples include AI co-pilots for search, summarization, and email
drafting. i.e., GitHub Copilot

Current Level – First level constrained autonomy. AI Agents can create, execute, and adapt
plans based on feedback, handling complex tasks with multiple reasoning cycles. Examples
Level 3 include reconciling a 100-page invoice against internal systems with human-like reasoning. i.e.,
Agentforce

While still theoretical today, they are technically feasible. AI Agents can self-
improve, modify their instructions, create new tools, and adapt to evolving
Level 4 tasks. Examples include invoice reconciliation agent that autonomously
adds vendors and improves accuracy.

Full AGI (artificial general intelligence) – capable

of independent reasoning, creativity, and solving
Level 5 novel problems beyond training. Examples
include digital knowledge worker handling
complex tasks with no oversight.

40
Source: sema4.ai “The Five Levels of Agentic Automation”

Morgan Stanley Research 49

M
Foundation

MORGAN STANLEY RESEARCH

AI Agents Are Rapidly Improving theAbility to Complete Long Tasks

Length of tasks that generalist frontier model agents can complete autonomously has been
Bottom Line doubling approximately every 7 months for the last 6 years.

In under a decade, AI Agents are expected to independently complete a large fraction of software tasks that currently take
humans days or weeks
Time it takes human professionals to complete task

Source: METR “Measuring AI Ability to Complete Long Tasks” 41

Note: Shaded region represents 95% confidence interval

50
M
Foundation

MORGAN STANLEY RESEARCH

Part III
Organizing the Agent Landscape

42

Morgan Stanley Research 51

M
Foundation

MORGAN STANLEY RESEARCH

System of Record (SoR) vs System of Engagement (SoE) AI Agents

SoR vs. SoE One framework for organizing the agent landscape is the distinction between System of Record
Ai Agents style agents from System of Engagement style agents.

SoR Agent SoE Agent

Role: Functional Supervisor

Automation Focus: Domain or Role Specific Workflow / Process-Oriented

Value Proposition: Labor Displacement VS System Utilization

More Direct - Automation of Indirect - Time Savings on

Quantification of ROI:
Role Broader Workflows
Difficulty of Harder - New Tech Plus Broader
Hard - New Tech Leaning Curve
Implementation: Integrations
Narrow - Defined by Role of Broad - Defined by Scope of
Market Opportunity:
Domain Workflows

43
Source: Morgan Stanley Research

52
M
Foundation

MORGAN STANLEY RESEARCH

System of Record (SoR) Agents – Role Specific Automation

System of Record style agents focus on automating functions related to specific domains such as
SoR Agents coding, customer service, sales lead generation, business intelligence, etc. Examples include:
GitHub Co-pilot, Workday, Salesforce Agentforce and HubSpot Prospecting agent

Agent SoR agents take a functional role in terms of their automation objectives. This means that the agents operate within a
Role particular business function or application-specific domain

Agent Focus of SoR agents is on specific role(s) within a particular domain such as customer service agents, software
Focus developers, business analysts, research assistants, sales development representatives etc.

Value
The value proposition for SoR agents centers on labor displacement or labor avoidance
Proposition

Quantifiable With a value proposition often tied to labor displacement/avoidance, directly ascribing cost savings should be easier
ROI? with SoR style agents

Time to
Given the ability to deliver hard, measurable ROI, the monetization timeline for SoR agents should be sooner and
Market generally ahead of SoE agents
Impact

44
Source: Morgan Stanley Research

Morgan Stanley Research 53

M
Foundation

MORGAN STANLEY RESEARCH

System of Engagement (SoE) Agents – Workflow / Process-Oriented Automation

Unlike System of Record agents, System of Engagement agents focus on managing and
SoE Agents automating workflow that span across different systems and applications. Examples include:
ServiceNow Now Assist, GitLab Workflows and Atlassian Rovo

Agent SoE agents take on a more supervisory / managerial role in that they manage a workflow across multiple applications
Role and functional systems

Agent Compared to SoR agents that focus on automating particular roles within a given domain, SoE agents focus on
Focus automating workflows or an entire business process

Value The value proposition for SoE agents is about enabling better utilization of enterprise resources as well as better
Proposition utilization of the software vendor’s solution

Quantifiable With a value proposition more associated with utilization than labor displacement, measuring ROI is generally more
ROI? difficult for SoE style agents than SoR agents

Time to
Given less quantifiable ROI, the monetization timeline for SoE agents should be later and generally after SoR style
Market agents
Impact

45
Source: Morgan Stanley Research

54
M
Foundation

MORGAN STANLEY RESEARCH

Organizing Across Two Vectors

As investors look to understand the competitive landscape, we suggest organizing across two
Bottom Line vectors, the first being Workflow versus Role Specific, inline with the industry discussion around
'Systems of Record' versus 'Systems of Engagement’.

System of Record vs. System of Engagement // Horizontal vs. Vertical

• System of Record agents take a functional role in terms of

their automation objectives. This means that the agents
operate within a particular business function or application-
specific domain.
• Examples include: GitHub Co-pilot, Workday,
Salesforce Agentforce and HubSpot Prospecting
agent
• System of Engagement agents take on a more supervisory /
managerial role in that they manage a workflow across
multiple applications and functional systems We expect an
analogous dynamic within agents
• Examples include: ServiceNow Now Assist, GitLab
Workflows and Atlassian Rovo

46
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 55

M
Foundation

MORGAN STANLEY RESEARCH

Organizing Across Two Vectors

As investors look to understand the competitive landscape, we suggest organizing across two
Bottom Line
vectors, the second vector would be horizontal versus vertical oriented solutions

System of Record vs. Workflow Process // Horizontal vs. Vertical

• Vertical-oriented solutions look to automate process within

one specific industry (i.e., Harvey.ai looks to automate legal
workflows)
• Horizontal-oriented agents look to automate processes
common across all industries (i.e., HR and software
development).
These organizing principles likely presents trade-offs between
time to market and overall size of the market opportunity.
• Well defined, role-specific solutions from a transaction-
oriented agent
• Should be easier to get up and running
• Specifically targets cost savings/revenue opportunity
for clear ROI
• Strong vertical orientation likely furthers both these
benefits.
• Workflow-oriented agents, with more horizontal use case
• Target broader and bigger opportunities over time
• Take more time to get built out and deployed.
• The Microsoft 365 Copilot solution (inc. agentic capabilities)
targets the >400m current commercial Office users,
• The Agentforce solution for ServiceCloud serves 3 million Call
Center representatives in the US (based on Bureau of Labor
Statistics data) and perhaps 12-15 million globally.
47
Source: Morgan Stanley Research

56
M
Foundation

MORGAN STANLEY RESEARCH

Organizing Across Two Vectors

Market Landscape

Source: Company Data, Morgan Stanley Research 48

Note: Positioning within quadrants not indicative of relative performance or competitive positioning

Morgan Stanley Research 57

M
Foundation

MORGAN STANLEY RESEARCH

Organizing Across Two Vectors

Market Landscape

49
Source: Company Data, Morgan Stanley Research

58
M
Foundation

MORGAN STANLEY RESEARCH

Classifying AI Agents By Reasoning Paradigms

Reasoning Another framework to organize AI Agents is by the different classes of reasoning paradigms they
can employ. A few of the popular reasoning paradigms include: Chain of Thought, ReAct,
Paradigms Reflexion and Tree of Thought

Chain of Thought (CoT) ReAct

Prompts the agent to think in a linear, step This style of prompting brings together
by step by fashion by generating an reasoning in the form of thoughts with
internal reasoning chain before producing actions. Unlike CoT, ReAct agents are able
the final answer. Chain of Thought to interact with tools and its environment to
reasoning does not use any external tools generate an answer

Reflexion Tree of Thought (ToT)

Extends ReAct style reasoning by adding Unlike the linear reasoning employed by
two key elements: self-reflection and long- ReAct agents, Tree of Thought agents
term memory. By adding these capabilities search through a tree of possible
these agents can improve and learn over reasoning steps and evaluate each path
time with each successive run. before choosing the optimal one

Sources: 1) Shunyu Yao, Et al. “ReAct: Synergizing Reasoning and Acting in Language Models”, 2) Sunyu Yao, Et al. “Tree of Thoughts: Deliberate Problem solving With Large Language Models; 50
3) Jason Wei, Et al. “Chain-of-Thought Elicits Reasoning in Large Language Models, 4) Noah Shinn, Et al. “Reflexion:Language Agents with verbal Reinforcement Learning”

Morgan Stanley Research 59

M
Foundation

MORGAN STANLEY RESEARCH

Chain of Thought Agents

What are A type of AI agent that employes linear, step-by-step reasoning (thoughts) before producing an
CoT Agents? output

Logic
Thought 1→ Thought 2 → Thought 3 → Final Answer
Structure

Advantages Simple and easy and easy to implement and good for math, logic puzzles and general reasoning tasks

Limitations Does not interact with its environment, cannot access tools and does not make any observations during reasoning

51
Source: Prompt Engineering Guide (https://www.promptingguide.ai/)

60
M
Foundation

MORGAN STANLEY RESEARCH

Tree of Thought Agents

What are AI agents that decompose tasks into a hierarchical structure of potential actions or routes,
enabling the agent to assess possible solutions and enhance decision-making by discarding less
ToT Agents? viable paths at each stage before reaching the ultimate solution.

Logic Agent generates multiple solution paths → Forms decision tree → Evaluates paths → Selects the Optimal Path at
Structure each step

Allows for planning and a structured evaluation of different potential solutions while eliminating unpromising solutions
Advantages at each step. Useful in tasks that require complex decision making and deliberate planning

Do not interact with the environment, complex to implement, require capable evaluators and is computationally
Limitations expensive

52
Source: Tree of Thoughts: Deliberate Problem Solving with Large Language Models; Shunyu Yao et al (2023)

Morgan Stanley Research 61

M
Foundation

MORGAN STANLEY RESEARCH

ReAct Agents
What are AI agents that integrate reasoning with action are defined by their capacity to engage with the
React environment through tools, their capability to make observations, and their ability to further
Agents? reason based on those observations.

Logic
Thought → Take an Action → Observe → Repeat
Structure

Can interact with the environment and access various tools, adapts based on feedback and good at use cases
Advantages involving interactive tasks such as coding, Q&A augmented by retrieval

Limitations Does not have persistent memory across interactions and does not learn from past experience

53
Source: Prompt Engineering Guide (https://www.promptingguide.ai/)

62
M
Foundation

MORGAN STANLEY RESEARCH

Reflexion Agents
What are AI agents that resemble ReAct agents but incorporate self-reflection and long-term memory.
Reflexion Following each trial, the agent evaluates its successes and failures, retains feedback, and applies
Agents? this knowledge to enhance future performance.

Logic
(Thought → Act → Observe) → (Self-reflect → Update Long-term Memory) → Future Trial
Structure

Can interact with the environment and access various tools, adapts based on feedback and good at use cases
Advantages involving interactive tasks such as coding, Q&A augmented by retrieval

Limitations Does not have persistent memory across interactions and does not learn from past experience

54
Source: Prompt Engineering Guide (https://www.promptingguide.ai/)

Morgan Stanley Research 63

M
Foundation

MORGAN STANLEY RESEARCH

The Emerging AI Agent OS

Key
The emerging AI Agent OS consists of a few core components: 1) the LLM, 2) Tools, 3) Access to
Elements of
Memory and 4) the ability to plan
the Agent OS

Memory
Long-term Memory
Short-term Memory

AI Agent
Tools Planning
Code Interpreter LLMs Chain of Thought
Calculator
Reflexion
Web Search
Self-critique
APIs
Open AI Llama Claude Deep Seek Subgoal Decomposition
Data Sources

Action

55
Source: Lilian Weng’s Lil’Log (https://lilianweng.github.io/)

64
M
Foundation

MORGAN STANLEY RESEARCH

Breaking Down the Agent OS

The Four Core Components
Of Agent OS

Foundational Every agent is powered by an AI or Large Language Model to take instructions, call tools, apply reasoning, execute
Models tasks, structure response and deliver outputs

Define the set of actions an AI agent can take and can include code interpreters, calculators, calendars, web search,
Tools enterprise search, public/private APIs and databases/data sources

Refers to the process used to acquire and store and retrieve information. There are two major categories of memory:
1) short-term and 2) long-term. Short-term memory (aka working memory) is enabled by approaches such as retrieval
Memory augmented generation (RAG) to provide the agent necessary information and context to execute the task at hand.
Long-term memory refers to the ability of the Agent to recall information over extended episodes usually by leveraging
a vector/operational data store

Refers to the ability of the agent to decompose the problem into subgoals to enable efficient execution of tasks.
Planning Furthermore, planning includes the ability to reflect and refine behavior through self-criticism, self-reflection and
learning from mistakes

56
Source: Lilian Weng’s Lil’Log (https://lilianweng.github.io/)

Morgan Stanley Research 65

M
Foundation

MORGAN STANLEY RESEARCH

Multi-agent Frameworks

What Is an Multi-agent AI frameworks is software that simplifies the process of building systems by
Agent providing high levels of abstraction including ready-made components to handle tasks such as
Framework? orchestration, memory management, tool usage and reasoning

Agent Frameworks Showing Significant Signs of Market Traction Today

A python-based, open source multi- Created by LangChain, LangGraph is an An open source programming framework
agent framework that is designed to open source framework that allows developed by Microsoft, Autogen allows
orchestrate agents in a collaborative developers to build complex agentic developers to build agents and enables
environment. Agents are defined by their applications using a graph-based agents to work together to accomplish
roles and collaborate intelligently to architecture to model and manage the tasks by replicating the structure of
achieve common objectives relationships involved in an AI workflow human teams and organizations

57
Source: LangGraph, CrewAI, Microsoft websites

66
M
Foundation

MORGAN STANLEY RESEARCH

Agent Protocols

What are
Protocols enable uniform standards for AI models as well as agents to connect to external
Agent
services and for fostering agent to agent communication and collaboration
Protocols?

Leading Agent Protocols

MCP A2A ACP

Developed by Anthropic, Model Context The Agent-to-Agent protocol is a cloud- An open agent orchestration standard
Protocol is a protocol that provides a based standard developed by Google developed by IBM & BeeAI, Agent
standard interface for LLMs to connect that enables AI agents to communicate Communication Protocol enables
to external tools, services and data such with each other across different structured communication, discovery
as databases, APIs and files. With systems. This allows agents from and coordination between AI agents.
MCP, developers can forego custom different vendors to collaborate and ACP differs from A2A in that it focuses
and repetitive logic to connect each coordinate action to execute a given on local runtime environment as
LLM to external services. workflow opposed to being cloud-based

58
Source: Edwin Lisowski Medium Blog (https://medium.com/@elisowski)

Morgan Stanley Research 67

M
Foundation

MORGAN STANLEY RESEARCH

Comparing and Contrasting The Major Agent Protocols

Model Context Protocol Agent Communication Agent-to-Agent Protocol
(MCP) Protocol (ACP) (A2A)
Developed by Anthropic IBM & BeeAI Google

Primary focus Enabling LLMs to get real-time Enabling local communication, Enabling cross-platform agent
context by creating a standard coordination and orchestration communication, collaboration
interface to tools, data & APIs between agents and orchestration
Architecture Client-server (host/server Decentralized, local runtime HTTP-based client/server with
model) Agent Cards

Scope Vertical integration: connection Agents running locally (say on Horizontal integration: agent-
to tools/services to AI models a desktop or single server) to-agent communication &
coordination
Ecosystem Support Wide acceptance across the Linux foundation, BeeAI and Over 50 companies have
ecosystem including Microsoft, IBM Research announced support including
Open AI, Google, AWS, Altassian, Salesforce, SAP,
Cloudflare, Paypal, MongoDB, MongoDB, LangCahin, Elastic,
Atlassian and Glean Datadog and Confluent
Security Model App-layer auth, OAuth2, Runtime sandboxing, private OAuth2, scoped endpoint
scoped APIs network security exposure

Best for LLM apps with external Edge AI, embedded systems, Multi-agent workflows across
data/tool needs offline agents platforms in the cloud

Use Case Connecting an LLM to internal On-device coordination of Distributed enterprise agents
APIs multiple small agents collaborating

59
Source: Edwin Lisowski Medium Blog (https://medium.com/@elisowski)

68
M
Foundation

MORGAN STANLEY RESEARCH

MCP: Unlocking a New Wave of Agentic Application Development

For AI agents to be effective, they need to take action in real-time. This means getting relevant
What’s the context, connecting to data and accessing a suite of tools. Prior to MCP, developers would have
Problem? to build custom integrations to connect to the necessary tools and services for each LLM they
were running

Solution MCP provides a universal standard for LLMs to connect apps, data and services without the need for coding

Just as personal computers utilize USB as a standard way to connect to a host of services (storage, chargers,
An Analogy headphones, monitors), AI models use the MCP protocol as a standard way to extend their knowledge and
capabilities

The emergence of MCP makes it easier to build AI agents that are useful and capable in real world environments
Implication helping to usher in new a wave of agentic applications

60
Source: Data Science Collective (https://medium.com/data-science-collective)

Morgan Stanley Research 69

M
Foundation

MORGAN STANLEY RESEARCH

Part IV
Market Landscape

61

70
M
Foundation

MORGAN STANLEY RESEARCH

Current Market Landscape

62
Source: Morgan Stanley Research

Morgan Stanley Research 71

M
Foundation

MORGAN STANLEY RESEARCH

Most Useful Domains

Customer Service Sales and Marketing Human Resources

Software Development and Operations IT Support Finance

63
Source: Morgan Stanley Research

72
M
Foundation

MORGAN STANLEY RESEARCH

Part V
Company Spotlights

64

Morgan Stanley Research 73

M
Foundation

MORGAN STANLEY RESEARCH

Hyperscalers – Microsoft
Microsoft Agents

• In March 2023, Microsoft announced Microsoft 365 Copilot

building the foundation for its agents, which plug into existing
business applications, security tools, and data management.
• Autonomous agents focused on key business functions (sales,
customer support, security, etc.) entered a paid preview
program on February 1, 2025, and pricing is based on the
complexity of responses and actions generated
• Microsoft’s agents are priced based on consumption,
which is tiered by the ‘level of autonomy’
• E.g. A hypothetical agent might answer questions from
customers on a website. Yesterday, it consumed 500 classic
answers and 2,000 generative answers. Therefore, it would
cost 4,500 messages, equivalent to $45 for that day.
• A hypothetical agent in Copilot Chat uses data stored in
Microsoft Graph to answer employee questions about HR
policies. Yesterday, the agent consumed 200 generative
answers and 200 tenant Graph grounding for messages.
Therefore, it would cost 6,400 messages or $64 for that day.
• A hypothetical autonomous agent responds to and routes
inbound sales orders from customers. Yesterday, it consumed
100 generative answers, 100 tenant Graph grounding for
messages, and 800 autonomous actions. Therefore, it would
cost 23,200 messages or $232 for that day.
• The consumption will be charged back to customers as MACC
credits (Microsoft Azure Consumption Commitment).
• Microsoft benefits from its comprehensive AI portfolio and its
breadth of Agent specific offerings, including: GitHub Copilot’s
asynchronous coding agent; Azure AI Foundry Agent Service
for orchestrating multiple specialized agents, Microsoft 365
Copilot Studio / Agent Build, Agents for Microsoft Teams, M365
Agents toolkit for Visual Studio, M365 Agents SDK, Data Agents
in Microsoft Fabric, SRE agents (among others)

65
Source: Company Data, Morgan Stanley Research

74
M
Foundation

MORGAN STANLEY RESEARCH

Hyperscalers – Amazon
Amazon Q

• An all-in one AI assistant, AMZN's Q spans across the business,

integrating with a user's AWS account, business intelligence tools,
contact center and business apps. Q spans across 40 popular data
sources including S3, Salesforce, Google Drive, Gmail, Microsoft
365 and others. After connecting these data sources, Q indexes all the
information, understanding core concepts and organizational
idiosyncrasies, allowing Q to provide comprehensive and personalized
answers specific to each user.
• Agentic Capabilities: Q's agentic capabilities also extend into
AWS services like Connect and QuickSight, enabling services
like real-time agent assist for contact centers or creation of
executive summaries of key business data for analysts. In
1Q25 Q launched a new agentic coding experience that can
execute complex workloads autonomously.
• Pricing: $20 and $25 a month, respectively, for the business
and builder tiers.

• Amazon Q Business: Users can leverage their company’s content,

data and systems by using Q to perform a series of tasks including;
search across company systems and data, app creation to accelerate
automation, actions in other applications (Jira, Salesoforce, PagerDuty,
etc.).
• Amazon Q Developer: Users can leverage Q developer for building,
operating and transforming software, helping developers with coding,
testing, deploying, trouble shooting and optimizing AWS resources.
AMZN has seen up to 40% improvement in developer productivity for
those who adopt Q Developer.

• AMZN developed Nova Act, a new AI model trained to perform actions

within a web browser that allows developers to break down complex
workflows into reliable atomic commands.

66
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 75

M
Foundation

MORGAN STANLEY RESEARCH

Hyperscalers – Google
Gemini

• GOOGL’s Gemini LLM is deeply integrated into a variety of enterprise

offerings, including Vertex AI, Google Workspace, Agentspace, and
Gemini Code Assist

• Key Capabilities:
TPU
• Vertex AI Agent Builder
• Easier Agent Creation: Enterprises can use an agent
development kit (ADK) to build conversational or task-
specific agents with no-code/low-code prompts that can
connect to their data, API, and workflows securely
• Agent2Agent (A2A) Protocol: Universal
communication standard that allows agents across
different ecosystems to communicate with each other,
allowing business agents to work together across
different tasks
• Gemini for Google Workspace
• Integration Across Drive: Gemini can link to an
enterprise’s Google Docs, Sheets, Slides, etc. and can
generate text, charts, emails, meeting notes etc.
• Agentspace
• Centralized Hub for Agents: Integrates with an
enterprise’s database and acts as a one-stop shop for
employees to search across enterprise data, use/create
agents, and generate video/image/text/code
• Gemini Code Assist
• AI Coding Assistant: Code generation, completion,
debugging that can integrate with enterprise databases

• Vertex AI Agent Builder and Gemini Code Assist are available currently
through Google Cloud; Gemini for Google Workspace is available in
Google Workspace Business and Enterprise plans, and Agentspace is
currently only available to select customers
67
Source: Company Data, Morgan Stanley Research

76
M
Foundation

MORGAN STANLEY RESEARCH

3rd Party Model Providers – OpenAI

Operator

• Launched in January 2025 as a research preview, Operator is an AI

Agent that combines advanced vision capabilities with browser
automation to perform web-based tasks independently.
• Operator is powered by a new model called Computer-Using Agent
(CUA), which combines 1) GPT-4o’s multimodal vision for analyzing
webpage screenshots, 2) reinforcement learning for decision-making,
and 3) virtual input simulation (mouse clicks, keyboard typing, scrolling).
• Operator interacts directly with graphical user interfaces (GUIs),
mimicking human browsing behavior, instead of using APIs. This allows
it to work across websites without requiring custom integrations.

• Key Capabilities:
• Task Automation: Handles repetitive web activities with 38%
success rate on OSWorld benchmarks (vs 72% human
baseline). Examples include grocery ordering, flight booking, etc.
• Workflow Personalization: Users can 1) save recurring task
sequences (i.e., weekly grocery orders), 2) set site-specific
preferences (i.e., airline seat selection), 3) run multiple
concurrent tasks through separate browser sessions.
• Security Protocols: Auto-switches to manual mode for
logins/payments, blocks screenshots during sensitive
interactions, and requires user confirmation before finalizing
orders/transactions.

• Currently available only to U.S.-based ChatGPT Pro subscribers, which

are paying $200/month.
• Token Pricing: GPT-4o charges $2.5/1M input tokens and $10/1M
output tokens.

68
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 77

M
Foundation

MORGAN STANLEY RESEARCH

3rd Party Model Providers – Anthropic

Computer Use

Pricing

• Launched in October 2024 as a public beta, Computer Use is an agentic

system that enables Claude 3.5 Sonnet models to interact with graphical
user interfaces (GUIs) using mouse movements, keyboard inputs, and
screen analysis – mimicking human-computer interactions.
• The agent combines vision-language models for screen interpretation
with reinforcement learning for decision-making, allowing it to navigate
software interfaces without API integration.

• Key Capabilities
• Task Automation and Workflow Execution: The agent
autonomously executes multi-step digital tasks, including 1) web
browsing, 2) application management, content creation, and
administrative workflows.
• Unlike traditional RPA, it adapts to unstructured
interfaces and dynamic workflows without predefined
scripts.
• Technical Implementation: The system works through a four-
stage loop, including 1) prompt analysis to interpret user
instructions, 2) chooses between integrated capabilities/tools like
terminals, text editors, or browser controls, 3) performs tasks
within a sandboxed Docker container or VM, taking screenshots
to verify outcomes, and 4) repeats steps until task completion or
a 10-iteration limit.

• Token Pricing: Claude 3.7 and 3.5 Sonnet charge $3/1M input tokens
and $15/1M output tokens.

69
Source: Company Data, Morgan Stanley Research

78
M
Foundation

MORGAN STANLEY RESEARCH

SaaS Vendors – Atlassian Corp (TEAM)

Rovo

Out-of-the-box Agents
• First introduced in April 2024 and launched to GA in October 2024,
Rovo is Atlassian’s first AI-powered SKU designed as an enterprise
search platform and virtual teammates (Agents) that can be called on
or created by any team member to collaborate and automate tasks.
• Rovo operates across three fundamental dimensions: Find (Rovo
Search), Learn (Rovo Chat), and Act (Rovo Agents).
• Rovo Search: Get relevant information from your Atlassian
tools (Cloud and Data Center), 3rd party SaaS applications,
and custom data sources. Use pre-built connectors that
seamlessly integrate with Jira, Confluence, Loom, and
Bitbucket, along with essential platforms like Google Drive,
Microsoft SharePoint, Figma, Slack, Microsoft Teams, and
GitHub.
• Rovo Chat: With deep understanding of your organization,
Rovo offers personalized answers and provides expert
advice. Answers questions about your team, projects, and
company.
• Rovo Agents: Out-of-the-Box Agents can easily step into
tasks like writing release notes, translating content into
multiple languages, grooming your backlog and more. Build
Custom Agents with no-code/low-code or the Forge
development platform. Also explore a diverse range of
Agents created by 3rd party partners on the Atlassian
Marketplace. Pricing
• Rovo for GitHub Copilot provides instant code • Initially introduced with a $20/user/month price tag
suggestions tailored to a team’s design rules, Jira • Now access is free for Premium and Enterprise customers,
tasks, and more all within the IDE.
• Coming to Standard soon.
• Rovo’s truly “Agentic” nature is evident by its ability to execute
complex tasks autonomously by leveraging AI models while • $5/month for non-Atlassian users
integrating seamlessly into existing workflows. Its purpose-built • Consumption limit in place to control costs, with
agents can perform tasks with minimal human intervention, while overconsumption triggering a bill – though not punitive
learning and adapting in real-time. • Also included as part of new Teamwork Collection bundle along with
Jira, Confluence, and Loom
70
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 79

M
Foundation

MORGAN STANLEY RESEARCH

SaaS Vendors – ServiceNow, Inc (NOW)

Pro Plus – AI Agents

• ServiceNow announced the general availability of its Yokohama Agentic Pricing Not Specified, Though Management
release in early March, introducing AI Agent Studio, Workflow Commentary Suggests a Value-Based Metering System
Data Fabric, and providing customers with teams of Agentic Commentary
preconfigured AI agents, enabling users to automate workflows "We have well over a 1,000 customers using our AI platform in Agentic AI. Well over 250 million in ACV. It's
and improve productivity levels Investor Day
the fastest growing product family in the history of ServiceNow , and it will continue to be that. And as Amit said,
it's a hybrid pricing model. And what that means is, when we delivered Pro Plus on day one, there was a seat-
• The company offers its solutions in three packages: 1) Standard, based license like we've always had, but there was also a consumption based that we built into that licensing
model. And that's super important as we move forward. "
2) Pro, and 3) Pro Plus, with access to AI Agents being offered
within the Pro Plus package. The Pro Plus offering provides "Q1 was another substantial acceleration for ServiceNow AI. The number of Pro Plus deals more than
customers with an initial set of agentic capacity with potential to 1Q25 Transcript quadrupled year-over-year, including 39 deals with three or more Now Assist products. Average ACV deal
sizes grew by one-third quarter-over-quarter as Pro Plus products were included in 15 of our top 20 deals. "
purchase / add further capacity should organizations exceed
initial usage levels
• AI Agent Studio: Provides users the ability to efficiently design Morgan Stanley TMT Conference
"The more you use it and the more it accommodates interactions with multiple systems across your
business or your network, the more we will make on the consumption-based pricing model. "
custom AI agents that integrate into their existing enterprise-wide
workflows and data. The solution leverages a no-code language-
based interface, enabling customers to build a team of AI agents "As the agents become increasingly productive, they will drive the consumption pricing meter. And that, of
course, will be in addition to the seat-based licensing foundation. You'll get both."
• Examples of pre-configured AI agents include: 4Q24 Transcript
"If they (customers) go beyond the capacity and the Assist, which is the metered capacity, if they go beyond
• Security Operations Expert that, they're still buying Assist packs. So it's not like one assist per time in terms of consumption."

• Autonomous Change Management

• Proactive Network Test and Repair Workflow Data Fabric Serves as a Key Piece of ServiceNow’s
• Operational Technology Knowledge Generation Agentic Architecture
• ITSM Incident Categorization
• Post-Incident Review Generation
• Workflow Data Fabric: A core element of ServiceNow’s Agentic
offering as it’s powered by the company’s automation engine and
serves as an integrated data layer that unifies enterprise data to
more efficiently process workflows and power AI agents with real-
time, secure access to data from a variety of sources
• Agentic AI is “The New Frontier” and ServiceNow is at the
Core as its preconfigured AI agents address a broad set of
complex workflows while its data connectivity (via Workflow Data
Fabric) provides unique differentiation vs peers as it provides AI
agents incrementally more data to leverage thereby making them
more impactful across enterprise employee workflows over time
71
Source: Company Data, Morgan Stanley Research. Note: Pre-configured agents are a small sample of others that are offered.

80
M
Foundation

MORGAN STANLEY RESEARCH

SaaS Vendors – Salesforce (CRM)

Agentforce
• Agentforce (General Availability in October 2024) is a suite of autonomous, Agentforce Work Interface
out-of-the-box agents, allowing customers to automate administrative
tasks while reorienting employee bandwidth to higher priority work.
• Service agents: automatically help a customer troubleshoot an
issue with personalized, branded advice based on their purchase
history and company knowledge documents
• SDR agents: send a personalized introductory email to a lead,
respond to a question on pricing, and access an account
executive’s calendar to schedule a follow-up call
• Marketing agents: generate campaigns, briefs, audience
segments, and emails based on your company's goals
• Ecommerce agents: manage websites and generate
personalized promotions
• The latest version, Agentforce 2dx (announced March 2025) expands
beyond reactive chat interfaces to enable proactive AI agents that work
behind the scenes without constant human oversight, with capabilities
including multimodal experiences and embedded AI in any workflow.
AgentExchange, a trusted marketplace and community where partners
can build and monetize agentic AI, will accelerate the deployment of AI
agents. Pricing
• Strong Early Customer Traction: As of F1Q26, Salesforce signed
>4,000 paid Agentforce deals, up from ~3k the previous quarter, with
Agentforce now crossing a $100M ARR run rate since going GA.
• Pricing: After initially being rolled out at a fixed price of $2 per call,
Salesforce recently added ‘Flex Credits” to the Agentforce pricing model, a
consumption-based model that ensures customers only pay for the actions
Agentforce performs. Credits are available for purchase in packs of 100k
at a $500 price point– each action consumes 20 credits (or $0.10). The
pay per action model better connects price to value and allows customers
to scale usage of Agentforce into additional workflows (beyond call
centers) at a more favorable price.
• The combination of broad application functionality, deep data sets to power
the AI models and an Agentic Computing architecture to take action
frames CRM's strong positioning for GenAI.
72
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 81

M
Foundation

MORGAN STANLEY RESEARCH

SaaS Vendors – Intuit (INTU)

Intuit AI Agents
• During Intuit's 3Q FY25 Earnings call, CEO Sasan Goodarzi introduced Quickbooks Interface
their new suite of specialized AI Agents, focusing on automating tasks
such as invoicing, payment tracking, transaction reconciliation, and project
management. These agents are designed to act proactively, offering
insights and executing tasks on behalf of users, streamlining operations for
SMBs.
• Why is Intuit a Wildcard? Given small businesses propensity to buy
bundled solutions, we like Intuit’s opportunity to upsell agentic capabilities
into their broad install base (10M+ small and mid-market businesses as of
Jan ‘25). That said, given the solution is still not generally available, and
we have not seen it in the marketplace (pricing dynamic still unknown), we
keep intuit in the wildcard bucket of our Agentic Winners Framework.
• Accounting Agent: keeps books clean and accurate by
automatically categorizing transactions, detecting and resolving
anomalies, and reconciling accounts—flagging what needs
approval or attention.
• Payments Agent: helps facilitate faster payments, and bills paid
on time by monitoring cash flow and optimizing invoicing and
payment collection. Offerings
• Finance Agent: analyzes financial data and automatically creates
forecasts, highlights variances, and pinpoints the source to help
you stay on track to your goals.
• Customer Agent: your clients can source leads from their inbox,
draft personalized email responses, and track every customer
opportunity in the sales cycle.
• Our Thoughts on Pricing:
• Pricing for Value: Intuit estimates SMBs value their time at about
$75 per hour
• Add-on module pricing: plan to introduce AI capabilities as add-
on modules in the coming months

73
Source: Company Data, Morgan Stanley Research

82
M
Foundation

MORGAN STANLEY RESEARCH

SaaS Vendors – Adobe (ADBE)

Adobe Experience Platform Agent Orchestrator & Purpose-built Agents
More than $3.5B AI-Influenced Revenue
• At the 2025 Summit conference, Adobe unveiled the Adobe
Experience Platform Agent Orchestrator, which enables
businesses to build, manage, and orchestrate AI agents from Adobe
and third-party ecosystems.
• Adobe released 10 purpose-built agents with a deep
understanding of customer data and ability to support multiagent
collaboration and adaptive reasoning. These agents include:
Account Qualification Agent, Audience Agent, Content Production
Agent*, Data Insights Agent, Data Engineering Agent,
Experimentation Agent, Journey Agent, Product Advisor Agent, Site
Optimization Agent*, Workflow Optimization Agent*
• Extended collaboration with Microsoft.
• Adobe Express Agent for Microsoft 365 Copilot enables
users to create high-quality images directly within Copilot,
PowerPoint and Word.
• Adobe Marketing Agent for Microsoft Copilot 365
empowers marketers to refine audience targeting, extract
actionable insights and improve cross-team collaboration. AI Agents incorporated to the AEP
• Brand Concierge: An agentic app that draws on a company’s
unique brand attributes and customer data to enhance brand loyalty
by engaging customers through multimodal interactions across text,
voice, and images.
• Pricing: The company has not disclosed specific pricing for these
agents. Their statements around general pricing strategy for AI
SKUs at the Summit in March 2025 suggest they will likely follow
their established pattern of tiered, value-based pricing approaches
for these new AI capabilities.
• Adobe is capitalizing on its domain expertise across content, data
and workflows to drive competitive advantage with GenAI. While
monetization takes time, the product road-map and positive partner
conversations provide comfort on Adobe's competitive moat.
*Included as part of the GenStudio.
74
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 83

M
Foundation

MORGAN STANLEY RESEARCH

SaaS Vendors – Workday (WDAY)

Illuminate

• At Workday’s Rising conference in September 2024, Workday “We monetize it through one of three ways. Number one, our agents, whether it's our
introduced Illuminate, which is their platform of AI products and Recruiter Agent or – we have a new agent that we've announced that's really an internal
models built on Workday’s HR and finance datasets. Illuminate is mobility agent for employees to start to think about how they move around. We have the
succession agent, we have a whole bunch of agents. We announced the policy agent. We
focused on three key areas: 1) Accelerating tasks with GenAI, 2)
announced a financial audit agent. So we're going to build our own agents. And that's a
Assisting users with Copilots like Workday Assistant, and 3) way for us to monetize the Workday platform.
Transforming business processes through role-based agents.
• Workday has rolled out multiple role-based agents so far: At the same time as others out there start to build agents, we've developed something as
part of the Agent System of Record an agent gateway. So when people want to get
• Recruiting Agent: Provides AI-powered talent sourcing and
access and onboard into the enterprise, a new digital agent or digital employee, we're
screening capabilities, boosting recruiting solution selling going to monetize this new gateway as people come through and get access to our
price by 150% platform. So it's another way for us to monetize it.
• Expenses Agent: Provides users with flexible ways to
submit and approve expenses, while giving leaders the And then the third way we're monetizing it is into an agent or the agent gateway. But it's
ability to set controls and analyze spend our products that include AI. An example is Extend Pro. Another example would be what
we're doing with something like talent optimization, which is a SKU, it's not an agent, but
• Succession Agent: Assists HR teams with succession it's 100% driven or through something like Evisort, which is a CLM solution that a 100% AI
planning by identifying and developing potential candidates driven as well. So we have many different ways to monetize AI going forward and we think
for key positions within the organization we're uniquely positioned.”
• Contracts Agent: Continuously analyzes contracts across
- Carl Eschenbach, CEO, at Morgan Stanley TMT Conference, March 2025
the enterprise, surfaces obligations and opportunities buried
in unstructured data, and drives business actions to capture
value and mitigate risk
• Payroll Agent: Identifies and updates invalid payroll data,
automates audit workflows, surfaces insights, recommends
fixes, monitors compliance, and delivers system updates
• Financial Auditing Agent: Increases efficiency during
audits by connecting complex business documents to
monitor transactions, reconcile balances, and review internal
controls, allowing audit firms to develop apps that connect
directly to their Workday customers
• Underlying these role-based agents is Workday’s Agent System of
Record, a centralized platform to manage an organization’s entire
fleet of AI agents (both from Workday and third parties). This system
helps organizations maintain control, governance, and visibility while
leveraging AI agents.
75
Source: Company Data, Morgan Stanley Research

84
M
Foundation

MORGAN STANLEY RESEARCH

SaaS Vendors – HubSpot (HUBS)

Breeze

• HubSpot introduced their Breeze platform, which delivers AI

AI Powers the HubSpot Customer Platform
functionality to customer-facing teams to boost productivity, scale
growth, and unlock actionable insights, in September at 2024’s
INBOUND Conference.
• HubSpot currently has four Breeze Agents rolled out in the market, all
still in Beta aside from Customer Agent (went GA with credit-based
pricing, for Pro and Enterprise tiers as of June 2nd), acting as AI-powered
experts designed to automate workflows from planning through to
execution:
• Content Agent: Utilizes CRM data to streamline content
creation for engaging and differentiated marketing campaigns. It
helps marketers quickly produce different types of content that
attract visitors and generate leads and drive conversion,
including high-quality landing pages, podcasts, case studies, and
blogs.
• Social Media Agent: AI-powered expert that enhances your
social media strategy, analyzing your social performance, Pricing
company details, and marketing best practices to develop a
tailored, multi-channel content strategy that amplifies your social
presence. After driving consistently strong adoption/utilization with agents,
• Prospecting Agent: The ‘always on’ agent drives pipeline HubSpot is moving to direct monetization as of June 2nd, with
generation 24/7 by researching and executing personalized Customer Agent now available via credit-based pricing (in
outreach strategies for ideal prospects within HubSpot's Smart Pro/Enterprise tiers).
CRM.
• Customer Agent: Engages with customers on websites in full
conversation and at any time to provide instant answers– the
agent utilizes customer-specific knowledge base articles, URLs,
help sites, and blogs to deliver accurate and personalized
responses to common customer issues.
• HubSpot is well positioned to address the agent opportunity given: 1)
deep data sets combining rich structured and unstructured data; 2) a
unified platform approach that provides context across data,
engagement (hubs), and action (AI-powered execution); and 3) a vibrant
agent ecosystem developed through the company’s Agent.ai network
76
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 85

M
Foundation

MORGAN STANLEY RESEARCH

SaaS Vendors – UiPath, Inc (PATH)

Agent Builder – UiPath Maestro (Orchestrator) – AutoPilot
UiPath’s Pre-Built Agentic Workflow Templates and Prompts

• UiPath’s product innovation remains healthy with AI-focused

solutions including Agent Builder, Agentic Orchestration, and
AutoPilot announced in 2024.
• The company has made further progress with recent
announcements of UiPath Platform for Agentic Automation, and
bi-directional integrations with Microsoft Copilot Studio and other
3rd party agents through UiPath Maestro.
• AutoPilot: Provides a set of AI-powered solutions that can
automate and streamline a variety of general and department-
specific tasks. Core capabilities include automation generation,
self-service and assisted experiences, cross-application
compatibility, data processing, contextual awareness, process
optimization, and compliance. The functionality is available in the
below UiPath products
• Developers: Studio, StudioX, Apps, and Studio Web
• Testers: Test Manager and Studio
• Everyone: Assistant
• Agent Builder: Announced at the company’s FORWARD user
conference in 2024, Agent Builder enables users to create agents
from scratch or from the company’s pre-configured agents'
catalog. The offering launched with “hundreds of customers” in
private preview in December and management noted thousands
of agents have been created generating >250,000 agent runs to
date (as of 1Q26)
• UiPath Maestro: Combines RPA, AI, and human collaboration in
a unified orchestration platform. Designed for complex, long-
running enterprise workflows, Maestro enables organizations to
model, execute, and optimize processes that span many systems
(175 average in large enterprises) while reducing implementation
timelines.

77
Source: Company Data, Morgan Stanley Research

86
M
Foundation

MORGAN STANLEY RESEARCH

SaaS Vendors – GitLab (GTLB)

GitLab Duo

• First introduced in May 2023 as part of the GitLab 16 launch and

released to general availability in August 2024, GitLab Duo is a
comprehensive suite of AI-powered tools integrated into GitLab’s
platform, designed to assist users throughout the software
development lifecycle, from code creation and security
assessment to collaboration and analytics.
• GitLab Duo distinguishes itself in the market in two core aspects
• Goes beyond code assistance, integrating across the
platform to enable greater automation across the SDLC.
• Operates with a privacy-first approach, emphasizing
that it does not use customers’ proprietary code as
training data. This is particularly important for enterprise
customers concerned about IP protection.
• GitLab has established strategic partnerships to power its AI
capabilities, leveraging Google’s Vertex AI for tuning foundation
models and Anthropic’s Claude for specific features like code
suggestions. GitLab selects the most suitable LLM for each
individual feature.
• GitLab Duo is currently offered in two tiers, Duo Pro which is GitLab Duo Pricing
largely an AI code assistant and Duo Enterprise which provides Duo Pro Duo Enterprise
summarization, security, and analytics capabilities. $19 per user/month Contact Sales
Available for Premium and Ultimate
• GitLab Duo Workflows – often described as an “agentic customers
Available for Ultimate customers
workflow system” represents the next evolution, moving beyond Features Include: Everything from Duo Pro, plus:
reactive, prompt-based AI assistants and functioning as an Organizational User Controls Summarization and Templating tools Advanced Troubleshooting
autonomous team member that actively contributes to software User permissions for AI capabilities Discussion summary Root Cause Analysis
development processes. Code Suggestions Merge request summary AI Analytics
Code generation Code review summary AI Impact and Productivity Reporting
• Duo Workflows actively monitors projects, anticipates
Code completion Security and Vulnerability tools Personalized GitLab Duo
issues, automatically identifies and resolves Available in many popular IDEs and
vulnerabilities, optimizes applications for performance, Vulnerability explanation Self-Hosted (optional)
supports >20 programming languages
and streamlines development environments. Chat Vulnerability resolution Model Personalization (planned)
• Currently in private beta stage. Code explanation Advanced Chat with
Test generation Merge Request context
Code refactoring Issue and Epic context
78
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 87

M
Foundation

MORGAN STANLEY RESEARCH

SaaS Vendors – Appian (APPN)

AI Skills

• Initially launched in May 2023 as part of the Appian 23.2 release, and significantly enhanced through its 24.2 release in 2024, Appian AI Skills
represents the company’s implementation of agentic AI within its low-code process automation platform. It exemplifies practical agentic AI,
balancing purpose-built enterprise processes automation with human oversight.
• Through its collaboration with AWS Bedrock, Appian allows users to leverage the latest Gen to:
• Incorporate AI agents into mission-critical processes.
• Train custom AI models.
• Build your own generative AI prompts in processes.
• Leverage the latest models while maintaining privacy.
• Automate common use cases like content generation and processing, PII extraction, and data extraction.

Pricing

Pricing is based on a token limit system

within Appian’s subscription tiers:
• Standard tier: 100M tokens/month
• Advanced tier: 200M tokens/month
• Premium tier: 500M tokens/month

The exact cost per token is not publicly

disclosed.
Appian’s plan pricing is customized and
requires contacting sales for a quote.

79
Source: Company Data, Morgan Stanley Research

88
M
Foundation

MORGAN STANLEY RESEARCH

SaaS Vendors – Palantir (PLTR)

Palantir AIP

• Palantir’s Artificial Intelligence Platform (AIP)

which today includes agent functionality was
first announced in April 2023.
• AI Agent Functionality: AIP enables users to
build and deploy AI Agents using customized
LLMs that interact with enterprise data and
integrate with the customers’ “ontology”
(digital twin) to improve the decision-making
capabilities of the agents. Palantir’s agent use
cases focus primarily on data heavy, analytics
use cases, but can be broadly customized for
different operational use cases.
• Integration: AIP integrates with Palantir’s
Foundry, leveraging existing data pipelines
and operational workflows to automate
processes. Based on recent demos and
customer conversations, most agent Pricing
workflows continue to have a human in the
loop rather than being fully autonomous.

• Pricing details are not publicized but our channel conversations

suggest that customers can in many cases get initial developer
access via pilots (free trials), while enterprise pricing is then
provided on a custom basis based on existing contracts.

80
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 89

M
Foundation

MORGAN STANLEY RESEARCH

SaaS Vendors – C3.ai (AI)

C3 AI Agents

• C3 AI’s AI agent functionality was announced

initially as part of the C3 Generative AI suite in
March 2023.
• AI Agent Functionality: C3 AI agents leverage
orchestration software to create and deploy multiple
specialized AI agents that autonomously retrieve,
analyze, and summarize enterprise data via a
natural language interface, driving interactive
decision support and workflow automation.
• Integration: Agent capabilities are built into the C3
AI Platform and integrate with different pre-built
enterprise AI applications, to extend existing
models and connectors.
Pricing

• C3 AI’s pricing is partly consumption-based with pilots starting at

$250,000 to launch an AI application, which is followed by charges
per vCPU/vGPU hour with volume discounts.
• The service is available via cloud marketplaces for self-service
deployment.

81
Source: Company Data, Morgan Stanley Research

90
M
Foundation

MORGAN STANLEY RESEARCH

SaaS Vendors – SAP (SAPG)

Joule

SAP's AI strategy centers around Joule, its generative AI copilot launched in Out-of-the-box Agents
September 2023. Joule is embedded throughout SAP's cloud enterprise
portfolio, allowing users to interact with business data using natural language
across applications from HR to finance, supply chain, and customer
experience.

Joule serves as an intelligent assistant that understands business context and

can perform tasks like analyzing financial data, optimizing supply chains, and
enhancing customer experiences. It provides personalized insights and
recommendations based on users' roles and permissions within the SAP
ecosystem.

In October 2024, SAP enhanced Joule with collaborative AI agents -

specialized AI systems that work together on complex business workflows.
These agents handle specific tasks across finance, service, and sales with
customers able to build their own agents alongside SAP's library. The
agents can automate processes like invoice matching, customer service
inquiries, and sales opportunity qualification.

The SAP Knowledge Graph, introduced in Q1 2025, provides deeper business

understanding by mapping relationships across SAP's data landscape. This
solution grounds AI in business semantics, connecting entities like
purchase orders and invoices while reducing the risk of inaccurate AI results.

For developers, SAP offers AI-powered tools including:

• Generative AI capabilities in SAP Build for code explanation
• ABAP Large Language Model for customizations
• Vector database features in SAP HANA Cloud

SAP delivered over 130 AI use cases at the end of 2024 and has committed to
delivering 400+ by end of 2025, focusing on measurable business outcomes
while maintaining principles of relevance, reliability, and responsibility.
82
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 91

M
Foundation

MORGAN STANLEY RESEARCH

Startups – Cognition Labs

Devin

Claimed Performance

• Launched in March 2024, Devin is an autonomous AI assistant

created by Cognition Labs, branded as the “first AI software
engineer” and designed to complete software development tasks
with minimal human intervention.
• Cognition Labs (also known as Cognition AI) is a startup founded by
a small team led by CEO Scott Wu, CTO Steven Hao, and CPO
Walden Yan.
• The company received $21 million in Series A funding led by
Peter Thiel’s Founders Fund – at $350 million valuation.
• Raised another $175M shortly after in a Series B round led
by Founders Fund – at $2 billion valuation.
• Recently the company secured additional funding
(undisclosed amount) valuing the company close to $4 billion
as reported by Bloomberg.

• Core Functionalities:
• Autonomous planning and execution of complex engineering
tasks
• Coding, debugging, and problem-solving with ability to recall
relevant context and fix mistakes Pricing
• Learning unfamiliar technologies by researching online
resources to complete tasks After an initial early access period, Devin became generally available
• Building and deploying applications end-to-end from simple in December 2024 with pricing starting at $500 per month,
text prompts • Subscription includes no seat limits, access to its Slack integration,
• Finding and fixing bugs in existing codebases independently IDE extension, and API, along with onboarding and support.
• Training and fine-tunning its own AI models when necessary • There is a pay-as-you-go component for additional ACUs or “Devin
Activity Units” at $2/ACU, up to a limit. Each ACU is roughly
equivalent to 15 minutes of active Devin work.

83
Source: Company Data, Morgan Stanley Research

92
M
Foundation

MORGAN STANLEY RESEARCH

Startups – Glean
Work AI Platform

• Founded in 2019 by former Google search engineers led by Arvind

Jain, Glean has evolved from an AI-powered enterprise search
solution to a comprehensive “Work AI” platform.
• Beginning with a $15 million Series A in 2019, Glean has secured
over $500 million in total investment, including a $200 million Series
D round in February 2024 at a $2.2 billion valuation and a $260
million Series E round in September 2024 that valued the company
at $4.6 billion.
• Glean’s core offering is its Work AI platform, which connects to and
understands enterprise data, enabling AI-powered search, content
generation, and task automation.

• Key Platform Components:

• AI-Powered Search: Uses vector search and LLMs to
enable natural language queries. Delivered personalized
results based on the user’s role, current work, and
collaborators.
• Glean Assist: AI assistant that allows users to ask
questions, find information, and perform advanced tasks. In Pricing
February 2025, Glean enhanced this with “multi-step
prompts”, enabling reasoning through complex tasks. For organizations, there is no public pricing structure or free plan,
• Glean Agents: Launched in February 2025, it provides a requiring potential customers to schedule a demo and discussion with
horizontal agent environment for building, deploying, their sales team for a customized quote.
orchestrating, and governing AI Agents.
• These agents can automate personal tasks and For Individuals, there are 5 pricing editions:
business processes, with access to both enterprise 1. $12/month for monthly access
data and internet data. 2. $129 for 1 year - $10.75/month
3. $245 for 2 years – $10.20/month
4. $348 for 3 years - $9.67/month
5. $439 for 4 years – $9.15/month
84
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 93

M
Foundation

MORGAN STANLEY RESEARCH

Startups – Sierra
Agent OS

• Founded in February 2023 by Bret Taylor (current chairman of OpenAI’s

BoD and previously Co-CEO of Salesforce) and Clay Bavor (previously
led Google Labs), Sierra is a conversational AI platform that enables
businesses to deploy sophisticated, brand-aligned AI agents.
• In October 2024, Sierra raised $175 million in a Series B round led by
Greenoaks Capital at a $4.5 billion valuation, more than quadrupling its
valuation since January 2024 when it raised $110 million from Sequoia
Capital and Benchmark at a ~$1 billion valuation.
• Sierra has reportedly crossed $20 million in annualized
revenue, per Reuters.
• Sierra targets the customer experience software market by addressing
systemic inefficiencies in traditional contact centers. The platform,
branded as Agent OS, combines LLMs with proprietary orchestration
layers to create AI agents capable of handling complex, multi-step
customer interactions.

• Key Platform Components:

• Agent Development Environment: Provides no-code tools for
defining Goals (reducing call handling time or increasing upsell
rates), Guardrails (prevent unauthorized actions or violations),
and Knowledge Grounding (integration with 100+ enterprise
systems via APIs).
• Multi-LLM Architecture: Combines foundation models from
OpenAI, Anthropic, and Meta. Primary LLMs handle conversation Pricing
flow, while secondary models verify accuracy and monitor for
hallucinations. Sierra doesn’t publicly disclose specific pricing tiers, instead the
• Voice Interaction Engine: Launched in October 2024, it company emphasizes an “outcome-based pricing” approach, which is
processes tone, pacing, and emotional inflection, naturally tied to tangible business impacts, such as a resolved support
handles conversational overlaps/interruptions, and analyses conversation, a saved cancellation, an upsell, a cross-sell. If the
sentiment in real-time to trigger de-escalation protocols. conversation isn’t resolved, in most cases there’s no charge.
• Action Execution Framework: Agents can initiate business
processes like subscription cancellations, order
returns/exchanges, and appointment rescheduling.
85
Source: Company Data, Morgan Stanley Research

94
M
Foundation

MORGAN STANLEY RESEARCH

Startups – Aisera
Aisera Agentic AI Platform

• Founded in 2017 by CEO Matt Thomas, Aisera is an enterprise AI-

driven service automation platform, delivering solutions that blend
generative AI with sophisticated workflow orchestration.
• Since its funding, Aisera has raised $180 million across multiple
funding rounds, though its valuation remains undisclosed.

Key Platform Components:

• Intentless Orchestration Engine: Enables natural language
interactions without predefined intent mappings. It acts autonomously
to understand, reason, and proactively fulfill requests through AI
agents.
• Multi-LLM Gateway: Employs a model routing system that selects
between Azure AI, OpenAI, Google Vertex AI, Llama 3, AWS
Bedrock, or proprietary AiseraLLM based on query complexity.
• AI Agent Development: Build AI Agents leveraging AiseraLLM,
task-specific agents, domain-specific LLMs, and AI Studios.
• LLM Studio: Integrate and execute any LLM of your choice
in your Gen AI application with Aisera;s Gateway (LLMOps)
• Prompt Studio: Create, test, and refine prompts to mitigate Pricing per Azure Marketplace
hallucinations and irrelevant outputs.
• Event Studio: Take actions from system events by
dynamically triggering action flows based on the priority,
sequence, and nature of the event.
• Workflow Studio: Create workflows using a low-code/no-
code interface, leveraging a library of 3K+ pre-built
automations.
• Copilot Studio: Create context-aware and time-saving skill
flows that complete tasks easily.
• Hyperflow Studio: Complete tasks by identifying which
APIs to call and dynamically orchestrating their execution in
sequence on the fly.

86
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 95

M
Foundation

MORGAN STANLEY RESEARCH

Startups – CrewAI
CrewAI
• Founded in 2023 by João Moura in São Paulo, Brazil, CrewAI is a
multi-agent AI platform providing a framework for building,
deploying, and managing AI agents, with both high-level simplicity
and precise low-level control, for creating autonomous AI agents
tailored to any scenario:
• CrewAI Crews: Optimize for autonomy and collaborative
intelligence, enabling you to create AI teams where each agent
has specific roles, tools, and goals
• CrewAI Flows: Enable granular, event-driven control, single
LLM calls for precise task orchestration and supports Crews
natively.
• CrewAI began as an open-source framework (still is), but after a
successful private beta, it launched an Enterprise edition in October
2024 after raising $18 million across two funding rounds including a
Series A led by Insight Partners. Valuation is unknown.

• Key Features and Capabilities:

• Role-Based Agents: Create specialized agents with defined
roles, expertise, and goals.
• Intelligent Collaboration: Agents work together, sharing
insights and coordinating tasks to achieve complex objectives.
CrewAI Pricing Tiers, Features, and Intended Users
• Flexible Tools: Equip agents with custom tools and APIs to
Free Basic Standard Pro Enterprise Ultra
interact with external services and data sources. Pricing Free $99/month $6,000/year $12,000/year $60,000/year $120,000/year
• Task Management: Define sequential or parallel workflows, 2 Crews, 100 5 Crews, 1K 10 Crews, 2K 50 Crews, 10K 100 Crews,
with agents automatically handling task dependencies. 1 Crew, 50 monthly monthly monthly monthly 500K monthly
• Event-Driven Orchestration: Define precise execution paths Features monthly task executions, executions, executions, executions, executions,
responding dynamically to events. executions total of 5 unlimited unlimited unlimited unlimited
developers seats seats seats seats
• Fine-Grained Control: Manage workflow states and Small
conditional execution securely and efficiently. Small Median to Large
Techies businesses
Intended businesses large organizations Fortune 500/
• Native Crew Integration: Effortlessly combine with Crews for Users
searching for
with limited
that are
businesses, and Enterprises
enhanced autonomy and intelligence. a new AI tool scaling to
workflows growing enterprises
median
• Deterministic Execution: Ensure predictable outcomes with
explicit control flow and error handling.
87
Source: Company Data, Morgan Stanley Research

96
M
Foundation

MORGAN STANLEY RESEARCH

Startups – LangChain
LangChain – LangSmith – LangGraph
• Founded in October 2022 by CEO Harrison Chase and co-founder Ankush
Gola, LangChain originated as an open-source project to facilitate LangChain Ecosystem
development of applications connected to LLMs. After gaining significant
traction, it incorporated in April 2023 and has since raised ~$35 million in
private funding.

• Key Platform Components:

• Model I/O: Interfaces with language models through standardized
abstractions for LLMs, chat models, and embedding models
• Prompt Management: Creates templated and dynamically selected
prompts with example selectors to improve model outputs
• Retrieval Framework: Enables information retrieval through
document loaders, text splitters, embeddings, and vector stores for
effective RAG
• Memory Systems: Implements chat message history and
conversation persistence between chain runs
• Chains: Provides common building block compositions for creating
workflows that combine multiple components
• Agent Development: Builds AI agents that can reason about issues,
break them down into dub-tasks, and choose appropriate tools to
complete goals
• Ecosystem:
• LangChain Libraries: Core open-source packages available in
Python and JavaScript with various integrations
• LangChain Expression Language (LCEL): Declarative syntax for
orchestrating components, introduced in Q3 2023
• LangServe: Deployment tool launched in Oct 2023 to host LCEL code
as production-ready APIs
• LangSmith: Developer platform for debugging, testing, evaluating,
and monitoring LLM applications with pricing starting at
$39/seat/month
• LangGraph Platform: Cloud service for deploying agentic
applications at scale with usage-based pricing ($0.001/node executed)
88
Source: Company Data, Morgan Stanley Research

Morgan Stanley Research 97

M
Foundation

MORGAN STANLEY RESEARCH

Startups – Anysphere
Cursor

• Founded in 2022 by four MIT classmates (Sualeh Asif, Arvid Lunnemark,

Aman Sanger, and Michael Truell), the company launched its flagship
product Cursor in 2023. Later that year it raised $8 million in seed funding
led by OpenAI’s Startup Fund
• Anysphere has raised over $1 billion in funding to date, including a $900
million funding round in May 2025 led by Thrive Capital at a $9 billion
valuation

• Cursor is an AI-powered integrated development environment (IDE) that

enhances developer productivity through 1) intelligent code generation, 2)
codebase understanding, 3) multi-model integration, 4) smart reviews, 5)
VS code compatibility, and 6) Agent Mode

• Agent Mode: Autonomous AI coding agent that independently explores,

plans, and executes complex codebase changes with full tools.
Capabilities include:
• Autonomous Operation: Independently explores your
codebase, identifies relevant files, and makes necessary changes
• Full Tool Access: Uses all available tools to search, edit, create
files, and run terminal commands
• Contextual Understanding: Builds a comprehensive
understanding of your project structure and dependencies
• Multi-Step Planning: Breaks complex tasks into manageable
steps and executes them in sequence

• Ecosystem & Adoption:

• Enterprise Adoption: Used by engineering teams at major tech
companies including Stripe, OpenAI, Spotify, and Uber
• Developer Impact: Reportedly crossed 1 billion lines of
generated code per day
• Acquisition: Acquired AI coding assistant Supermaven in
November 2024 to expand capabilities
89
Source: Company Data, Morgan Stanley Research

98
M
Foundation

Morgan Stanley Research 99

M
Foundation

Disclosure Section
The information and opinions in Morgan Stanley Research were prepared by Morgan Stanley & Co. LLC, and/or Morgan Stanley C.T.V.M. S.A., and/or Morgan Stanley Mexico, Casa de Bolsa,
S.A. de C.V., and/or Morgan Stanley Canada Limited. As used in this disclosure section, "Morgan Stanley" includes Morgan Stanley & Co. LLC, Morgan Stanley C.T.V.M. S.A., Morgan Stanley Mexico,
Casa de Bolsa, S.A. de C.V., Morgan Stanley Canada Limited and their affiliates as necessary.
For important disclosures, stock price charts and equity rating histories regarding companies that are the subject of this report, please see the Morgan Stanley Research Disclosure Website
at www.morganstanley.com/researchdisclosures, or contact your investment representative or Morgan Stanley Research at 1585 Broadway, (Attention: Research Management), New York, NY,
10036 USA.
For valuation methodology and risks associated with any recommendation, rating or price target referenced in this research report, please contact the Client Support Team as follows: US/Canada
+1 800 303-2495; Hong Kong +852 2848-5999; Latin America +1 718 754-5444 (U.S.); London +44 (0)20-7425-8169; Singapore +65 6834-6860; Sydney +61 (0)2-9770-1505; Tokyo +81
(0)3-6836-9000. Alternatively you may contact your investment representative or Morgan Stanley Research at 1585 Broadway, (Attention: Research Management), New York, NY 10036 USA.

Analyst Certification
The following analysts hereby certify that their views about the companies and their securities discussed in this report are accurately expressed and that they have not received and will not
receive direct or indirect compensation in exchange for expressing specific recommendations or views in this report: Josh Baer, CFA; Brian Nowak, CFA; Elizabeth Porter, CFA; Chris Quintero;
Sanjit K Singh; Keith Weiss, CFA.
.

Global Research Conflict Management Policy

Morgan Stanley Research has been published in accordance with our conflict management policy, which is available at www.morganstanley.com/institutional/research/conflictpolicies. A
Portuguese version of the policy can be found at www.morganstanley.com.br

Important Regulatory Disclosures on Subject Companies

The analyst or strategist (or a household member) identified below owns the following securities (or related derivatives): Josh Baer, CFA - Amazon.com Inc(common or preferred stock), Peloton
Interactive, Inc.(common or preferred stock), Roblox Corporation(common or preferred stock), Uber Technologies Inc(common or preferred stock); Brian Nowak, CFA - Cloudflare Inc(common
or preferred stock); Elizabeth Porter, CFA - Microsoft(common or preferred stock); Chris Quintero - Expedia Inc.(common or preferred stock); Oscar R Saavedra - Alphabet Inc.(common or
preferred stock), Amazon.com Inc(common or preferred stock); Sanjit K Singh - Alphabet Inc.(common or preferred stock), Pinterest Inc(common or preferred stock); Keith Weiss, CFA - Uber
Technologies Inc(common or preferred stock).
As of April 30, 2025, Morgan Stanley beneficially owned 1% or more of a class of common equity securities of the following companies covered in Morgan Stanley Research: Adobe Inc., Airbnb
Inc, Akamai Technologies, Inc., Alphabet Inc., Amazon.com Inc, Appian Corp, AppLovin Corp, Asana Inc, Atlassian Corporation PLC, Autodesk, BILL Holdings Inc, Blackline Inc, Booking Holdings
Inc, Box Inc, Bumble Inc., C3.ai, CCC Intelligent Solutions Holdings Inc, Chewy Inc, Cloudflare Inc, Confluent, Inc., Couchbase, Inc., Coursera, Inc., Criteo SA, CrowdStrike Holdings Inc, CyberArk
Software Ltd, Datadog, Inc., DigitalOcean Holdings Inc, DocuSign Inc, DoorDash Inc, DoubleVerify Holdings Inc, E2open Parent Holdings Inc, eBay Inc, Elastic NV, Electronic Arts Inc, Etsy Inc,
Expedia Inc., Five9 Inc, Fortinet Inc., Gen Digital Inc., GitLab Inc, GoDaddy Inc, Intuit, Jamf Holding Corp, Klaviyo, Inc, LegalZoom.com Inc, Lightspeed POS Inc., Liveramp Holdings Inc, Lyft Inc, Match
Group Inc, Meta Platforms Inc, Microsoft, MongoDB Inc, NICE Ltd., Okta, Inc., Opendoor Technologies Inc, Oracle Corporation, PagerDuty, Inc., Palantir Technologies Inc., Palo Alto Networks
Inc, Peloton Interactive, Inc., Pinterest Inc, Qualys Inc, Rapid7 Inc, Reddit Inc, Revolve Group Inc, RingCentral Inc, Roblox Corporation, Salesforce, Inc., Samsara Inc, Semrush Holdings Inc -A,
ServiceNow Inc, Shopify Inc, Shutterstock Inc, Snowflake Inc., Sprinklr Inc, Sprout Social Inc, Take-Two Interactive Software, Tenable Holdings Inc, Toast, Inc., Trade Desk Inc, Twilio Inc, Uber
Technologies Inc, UiPath Inc, Varonis Systems, Inc., Vertex Inc., Webtoon Entertainment Inc, Workday Inc, WW International Inc, Yelp Inc.
Within the last 12 months, Morgan Stanley managed or co-managed a public offering (or 144A offering) of securities of Akamai Technologies, Inc., Alphabet Inc., AppLovin Corp, BILL Holdings
Inc, Booking Holdings Inc, Box Inc, Chewy Inc, CoreWeave, Datadog, Inc., DoorDash Inc, Klaviyo, Inc, Meta Platforms Inc, OneStream Inc, Sabre Corp, SailPoint Inc, ServiceTitan Inc, Uber
Technologies Inc, Unity Software Inc, Varonis Systems, Inc., Webtoon Entertainment Inc, Zeta Global Holdings Corp.
Within the last 12 months, Morgan Stanley has received compensation for investment banking services from Akamai Technologies, Inc., Alphabet Inc., Amazon.com Inc, AppLovin Corp, Autodesk,
BILL Holdings Inc, Booking Holdings Inc, Box Inc, CCC Intelligent Solutions Holdings Inc, Chewy Inc, CoreWeave, CyberArk Software Ltd, Datadog, Inc., DigitalOcean Holdings Inc, DocuSign Inc,
DoorDash Inc, Klaviyo, Inc, Meta Platforms Inc, OneStream Inc, Qualys Inc, Sabre Corp, SailPoint Inc, Salesforce, Inc., ServiceTitan Inc, Toast, Inc., Uber Technologies Inc, Udemy Inc, Varonis Systems,
Inc., Webtoon Entertainment Inc, Zeta Global Holdings Corp.
In the next 3 months, Morgan Stanley expects to receive or intends to seek compensation for investment banking services from Adobe Inc., Airbnb Inc, Akamai Technologies, Inc., Alphabet Inc.,
Amazon.com Inc, Amplitude Inc., Appian Corp, AppLovin Corp, Asana Inc, Atlassian Corporation PLC, Autodesk, BigCommerce Holdings, Inc., BILL Holdings Inc, Blackline Inc, Booking Holdings
Inc, Box Inc, Bumble Inc., C3.ai, CCC Intelligent Solutions Holdings Inc, Check Point Software Technologies Ltd., Chewy Inc, Cloudflare Inc, Confluent, Inc., CoreWeave, Couchbase, Inc., Coursera,
Inc., Criteo SA, CrowdStrike Holdings Inc, CyberArk Software Ltd, Datadog, Inc., DigitalOcean Holdings Inc, Docebo Inc., DocuSign Inc, DoorDash Inc, DoubleVerify Holdings Inc, Duolingo Inc,
eBay Inc, Elastic NV, Electronic Arts Inc, Etsy Inc, Expedia Inc., FIGS, Inc., Five9 Inc, Fortinet Inc., Freshworks Inc, Gen Digital Inc., GitLab Inc, GoDaddy Inc, HubSpot, Inc., Instacart, Integral Ad Science
Holding Corp., Intuit, Jamf Holding Corp, JFrog Ltd., Klaviyo, Inc, LegalZoom.com Inc, Lightspeed POS Inc., Liveramp Holdings Inc, Lyft Inc, Match Group Inc, Meta Platforms Inc, Microsoft, MongoDB
Inc, Nextdoor Holdings Inc, NICE Ltd., Okta, Inc., OneStream Inc, Opendoor Technologies Inc, Oracle Corporation, PagerDuty, Inc., Palantir Technologies Inc., Palo Alto Networks Inc, Peloton
Interactive, Inc., Pinterest Inc, Playtika Holding Corp, Qualys Inc, Rapid7 Inc, Reddit Inc, Revolve Group Inc, RingCentral Inc, Roblox Corporation, Sabre Corp, SailPoint Inc, Salesforce, Inc., Samsara
Inc, Semrush Holdings Inc -A, SentinelOne, Inc., ServiceNow Inc, ServiceTitan Inc, Shopify Inc, Shutterstock Inc, Snap Inc., Snowflake Inc., Sprinklr Inc, Sprout Social Inc, Take-Two Interactive
Software, Tenable Holdings Inc, Toast, Inc., Trade Desk Inc, Twilio Inc, Uber Technologies Inc, Udemy Inc, UiPath Inc, Unity Software Inc, Varonis Systems, Inc., Vertex Inc., Webtoon Entertainment
Inc, Wix.Com Ltd, Workday Inc, WW International Inc, Yelp Inc, Zeta Global Holdings Corp, Zillow Group Inc, Zoom Video Communications Inc, Zscaler Inc.
Within the last 12 months, Morgan Stanley has received compensation for products and services other than investment banking services from 8x8 Inc, Adobe Inc., Airbnb Inc, Akamai Technologies,
Inc., Alphabet Inc., Amazon.com Inc, AppLovin Corp, Asana Inc, Atlassian Corporation PLC, Autodesk, Blackline Inc, Booking Holdings Inc, Box Inc, CCC Intelligent Solutions Holdings Inc, Check
Point Software Technologies Ltd., Chewy Inc, Cloudflare Inc, Confluent, Inc., CoreWeave, DigitalOcean Holdings Inc, DocuSign Inc, DoorDash Inc, Dynatrace Inc, eBay Inc, Electronic Arts Inc,
Expedia Inc., Five9 Inc, Fortinet Inc., Freshworks Inc, Gen Digital Inc., Integral Ad Science Holding Corp., Intuit, Jamf Holding Corp, LegalZoom.com Inc, Match Group Inc, Meta Platforms Inc,
Microsoft, NICE Ltd., OneStream Inc, Oracle Corporation, Palantir Technologies Inc., Palo Alto Networks Inc, Peloton Interactive, Inc., Pinterest Inc, Playtika Holding Corp, Reddit Inc, Revolve
Group Inc, RingCentral Inc, Sabre Corp, SailPoint Inc, Salesforce, Inc., ServiceTitan Inc, Snap Inc., Snowflake Inc., Take-Two Interactive Software, Tenable Holdings Inc, Toast, Inc., Uber Technologies
Inc, Udemy Inc, UiPath Inc, Unity Software Inc, Workday Inc, WW International Inc, Zeta Global Holdings Corp, Zoom Video Communications Inc, Zscaler Inc.
Within the last 12 months, Morgan Stanley has provided or is providing investment banking services to, or has an investment banking client relationship with, the following company: Adobe

100
M
Foundation

Inc., Airbnb Inc, Akamai Technologies, Inc., Alphabet Inc., Amazon.com Inc, Amplitude Inc., Appian Corp, AppLovin Corp, Asana Inc, Atlassian Corporation PLC, Autodesk, BigCommerce Holdings,
Inc., BILL Holdings Inc, Blackline Inc, Booking Holdings Inc, Box Inc, Bumble Inc., C3.ai, CCC Intelligent Solutions Holdings Inc, Check Point Software Technologies Ltd., Chewy Inc, Cloudflare Inc,
Confluent, Inc., CoreWeave, Couchbase, Inc., Coursera, Inc., Criteo SA, CrowdStrike Holdings Inc, CyberArk Software Ltd, Datadog, Inc., DigitalOcean Holdings Inc, Docebo Inc., DocuSign Inc,
DoorDash Inc, DoubleVerify Holdings Inc, Duolingo Inc, eBay Inc, Elastic NV, Electronic Arts Inc, Etsy Inc, Expedia Inc., FIGS, Inc., Five9 Inc, Fortinet Inc., Freshworks Inc, Gen Digital Inc., GitLab
Inc, GoDaddy Inc, HubSpot, Inc., Instacart, Integral Ad Science Holding Corp., Intuit, Jamf Holding Corp, JFrog Ltd., Klaviyo, Inc, LegalZoom.com Inc, Lightspeed POS Inc., Liveramp Holdings Inc,
Lyft Inc, Match Group Inc, Meta Platforms Inc, Microsoft, MongoDB Inc, Nextdoor Holdings Inc, NICE Ltd., Okta, Inc., OneStream Inc, Opendoor Technologies Inc, Oracle Corporation, PagerDuty,
Inc., Palantir Technologies Inc., Palo Alto Networks Inc, Peloton Interactive, Inc., Pinterest Inc, Playtika Holding Corp, Qualys Inc, Rapid7 Inc, Reddit Inc, Revolve Group Inc, RingCentral Inc, Roblox
Corporation, Sabre Corp, SailPoint Inc, Salesforce, Inc., Samsara Inc, Semrush Holdings Inc -A, SentinelOne, Inc., ServiceNow Inc, ServiceTitan Inc, Shopify Inc, Shutterstock Inc, Snap Inc., Snowflake
Inc., Sprinklr Inc, Sprout Social Inc, Take-Two Interactive Software, Tenable Holdings Inc, Toast, Inc., Trade Desk Inc, Twilio Inc, Uber Technologies Inc, Udemy Inc, UiPath Inc, Unity Software Inc,
Varonis Systems, Inc., Vertex Inc., Webtoon Entertainment Inc, Wix.Com Ltd, Workday Inc, WW International Inc, Yelp Inc, Zeta Global Holdings Corp, Zillow Group Inc, Zoom Video
Communications Inc, Zscaler Inc.
Within the last 12 months, Morgan Stanley has either provided or is providing non-investment banking, securities-related services to and/or in the past has entered into an agreement to provide
services or has a client relationship with the following company: 8x8 Inc, Adobe Inc., Airbnb Inc, Akamai Technologies, Inc., Alphabet Inc., Amazon.com Inc, AppLovin Corp, Asana Inc, Atlassian
Corporation PLC, Autodesk, BigCommerce Holdings, Inc., Blackline Inc, Booking Holdings Inc, Box Inc, CCC Intelligent Solutions Holdings Inc, Check Point Software Technologies Ltd., Chewy
Inc, Cloudflare Inc, Confluent, Inc., CoreWeave, CyberArk Software Ltd, Datadog, Inc., DigitalOcean Holdings Inc, DocuSign Inc, DoorDash Inc, Dynatrace Inc, eBay Inc, Electronic Arts Inc, Etsy
Inc, Expedia Inc., Five9 Inc, Fortinet Inc., Freshworks Inc, Gen Digital Inc., GoDaddy Inc, HubSpot, Inc., Integral Ad Science Holding Corp., Intuit, Jamf Holding Corp, LegalZoom.com Inc, Match Group
Inc, Meta Platforms Inc, Microsoft, MongoDB Inc, NICE Ltd., OneStream Inc, Opendoor Technologies Inc, Oracle Corporation, PagerDuty, Inc., Palantir Technologies Inc., Palo Alto Networks Inc,
Peloton Interactive, Inc., Pinterest Inc, Playtika Holding Corp, Qualys Inc, Reddit Inc, Revolve Group Inc, RingCentral Inc, Sabre Corp, SailPoint Inc, Salesforce, Inc., ServiceNow Inc, ServiceTitan
Inc, Shopify Inc, Snap Inc., Snowflake Inc., Take-Two Interactive Software, Tenable Holdings Inc, Toast, Inc., Twilio Inc, Uber Technologies Inc, Udemy Inc, UiPath Inc, Unity Software Inc, Varonis
Systems, Inc., Wix.Com Ltd, Workday Inc, WW International Inc, Zeta Global Holdings Corp, Zillow Group Inc, Zoom Video Communications Inc, Zscaler Inc.
An employee, director or consultant of Morgan Stanley is a director of Alphabet Inc., eBay Inc, Elastic NV, Tenable Holdings Inc. This person is not a research analyst or a member of a research
analyst's household.
Morgan Stanley & Co. LLC makes a market in the securities of 8x8 Inc, Adobe Inc., Airbnb Inc, Akamai Technologies, Inc., Alphabet Inc., Amazon.com Inc, Amplitude Inc., Appian Corp, AppLovin
Corp, Asana Inc, Atlassian Corporation PLC, Autodesk, BigCommerce Holdings, Inc., Blackline Inc, Booking Holdings Inc, Box Inc, Bumble Inc., C3.ai, CCC Intelligent Solutions Holdings Inc, Check
Point Software Technologies Ltd., Compass, Inc., Confluent, Inc., CoreWeave, Couchbase, Inc., Coursera, Inc., Criteo SA, CyberArk Software Ltd, Datadog, Inc., DigitalOcean Holdings Inc, Docebo
Inc., DocuSign Inc, DoubleVerify Holdings Inc, Duolingo Inc, Dynatrace Inc, E2open Parent Holdings Inc, eBay Inc, Elastic NV, Electronic Arts Inc, Etsy Inc, Expedia Inc., FIGS, Inc., Five9 Inc, Fortinet
Inc., Freshworks Inc, Gen Digital Inc., GoDaddy Inc, Integral Ad Science Holding Corp., Intuit, Jamf Holding Corp, JFrog Ltd., Karooooo Ltd, Klaviyo, Inc, LegalZoom.com Inc, Liveramp Holdings Inc,
Match Group Inc, Meta Platforms Inc, Microsoft, Nextdoor Holdings Inc, OneStream Inc, Opendoor Technologies Inc, Oracle Corporation, PagerDuty, Inc., Palo Alto Networks Inc, Playtika Holding
Corp, Qualys Inc, Rapid7 Inc, Revolve Group Inc, RingCentral Inc, Sabre Corp, SailPoint Inc, Salesforce, Inc., Semrush Holdings Inc -A, ServiceNow Inc, ServiceTitan Inc, Shopify Inc, Shutterstock
Inc, Sprinklr Inc, Sprout Social Inc, Take-Two Interactive Software, Tenable Holdings Inc, Udemy Inc, Varonis Systems, Inc., Vertex Inc., Wix.Com Ltd, Workday Inc, Yelp Inc, Zeta Global Holdings
Corp, Zillow Group Inc, Zoom Video Communications Inc, ZoomInfo Technologies Inc, Zscaler Inc.
The equity research analysts or strategists principally responsible for the preparation of Morgan Stanley Research have received compensation based upon various factors, including quality
of research, investor client feedback, stock picking, competitive factors, firm revenues and overall investment banking revenues. Equity Research analysts' or strategists' compensation is not
linked to investment banking or capital markets transactions performed by Morgan Stanley or the profitability or revenues of particular trading desks.
Morgan Stanley and its affiliates do business that relates to companies/instruments covered in Morgan Stanley Research, including market making, providing liquidity, fund management,
commercial banking, extension of credit, investment services and investment banking. Morgan Stanley sells to and buys from customers the securities/instruments of companies covered in
Morgan Stanley Research on a principal basis. Morgan Stanley may have a position in the debt of the Company or instruments discussed in this report. Morgan Stanley trades or may trade
as principal in the debt securities (or in related derivatives) that are the subject of the debt research report.
Certain disclosures listed above are also for compliance with applicable regulations in non-US jurisdictions.

STOCK RATINGS
Morgan Stanley uses a relative rating system using terms such as Overweight, Equal-weight, Not-Rated or Underweight (see definitions below). Morgan Stanley does not assign ratings of Buy,
Hold or Sell to the stocks we cover. Overweight, Equal-weight, Not-Rated and Underweight are not the equivalent of buy, hold and sell. Investors should carefully read the definitions of all
ratings used in Morgan Stanley Research. In addition, since Morgan Stanley Research contains more complete information concerning the analyst's views, investors should carefully read Morgan
Stanley Research, in its entirety, and not infer the contents from the rating alone. In any case, ratings (or research) should not be used or relied upon as investment advice. An investor's decision
to buy or sell a stock should depend on individual circumstances (such as the investor's existing holdings) and other considerations.

Global Stock Ratings Distribution

(as of May 31, 2025)
The Stock Ratings described below apply to Morgan Stanley's Fundamental Equity Research and do not apply to Debt Research produced by the Firm.
For disclosure purposes only (in accordance with FINRA requirements), we include the category headings of Buy, Hold, and Sell alongside our ratings of Overweight, Equal-weight, Not-Rated
and Underweight. Morgan Stanley does not assign ratings of Buy, Hold or Sell to the stocks we cover. Overweight, Equal-weight, Not-Rated and Underweight are not the equivalent of buy,
hold, and sell but represent recommended relative weightings (see definitions below). To satisfy regulatory requirements, we correspond Overweight, our most positive stock rating, with a
buy recommendation; we correspond Equal-weight and Not-Rated to hold and Underweight to sell recommendations, respectively.

Morgan Stanley Research 101

M
Foundation

Other Material Investment Services

Coverage Universe Investment Banking Clients (IBC)
Clients (MISC)

Stock Rating % of Rating % of Total Other

Count % of Total Count % of Total IBC Count
Category Category MISC

Overweight/Buy 1493 40% 379 46% 25% 698 41%

Equal-weight/Hold 1650 44% 372 45% 23% 782 46%

Not-Rated/Hold 4 0% 0 0% 0% 2 0%

Underweight/Sell 602 16% 74 9% 12% 235 14%

Total 3,749 825 1717

Data include common stock and ADRs currently assigned ratings. Investment Banking Clients are companies from whom Morgan Stanley received investment banking compensation in the
last 12 months. Due to rounding off of decimals, the percentages provided in the "% of total" column may not add up to exactly 100 percent.

Analyst Stock Ratings

Overweight (O). The stock's total return is expected to exceed the average total return of the analyst's industry (or industry team's) coverage universe, on a risk-adjusted basis, over the next
12-18 months.
Equal-weight (E). The stock's total return is expected to be in line with the average total return of the analyst's industry (or industry team's) coverage universe, on a risk-adjusted basis, over
the next 12-18 months.
Not-Rated (NR). Currently the analyst does not have adequate conviction about the stock's total return relative to the average total return of the analyst's industry (or industry team's) coverage
universe, on a risk-adjusted basis, over the next 12-18 months.
Underweight (U). The stock's total return is expected to be below the average total return of the analyst's industry (or industry team's) coverage universe, on a risk-adjusted basis, over the next
12-18 months.
Unless otherwise specified, the time frame for price targets included in Morgan Stanley Research is 12 to 18 months.

Analyst Industry Views

Attractive (A): The analyst expects the performance of his or her industry coverage universe over the next 12-18 months to be attractive vs. the relevant broad market benchmark, as indicated
below.
In-Line (I): The analyst expects the performance of his or her industry coverage universe over the next 12-18 months to be in line with the relevant broad market benchmark, as indicated below.
Cautious (C): The analyst views the performance of his or her industry coverage universe over the next 12-18 months with caution vs. the relevant broad market benchmark, as indicated below.
Benchmarks for each region are as follows: North America - S&P 500; Latin America - relevant MSCI country index or MSCI Latin America Index; Europe - MSCI Europe; Japan - TOPIX; Asia -
relevant MSCI country index or MSCI sub-regional index or MSCI AC Asia Pacific ex Japan Index.

Important Disclosures for Morgan Stanley Smith Barney LLC Customers

Important disclosures regarding the relationship between the companies that are the subject of Morgan Stanley Research and Morgan Stanley Smith Barney LLC or Morgan Stanley or any
of their affiliates, are available on the Morgan Stanley Wealth Management disclosure website at www.morganstanley.com/online/researchdisclosures. For Morgan Stanley specific disclosures,
you may refer to www.morganstanley.com/researchdisclosures.
Each Morgan Stanley research report is reviewed and approved on behalf of Morgan Stanley Smith Barney LLC. This review and approval is conducted by the same person who reviews the
research report on behalf of Morgan Stanley. This could create a conflict of interest.

Other Important Disclosures

A member of Research who had or could have had access to the research prior to completion owns securities (or related derivatives) in the Atlassian Corporation PLC, Microsoft, Uber
Technologies Inc, Workday Inc. This person is not a research analyst or a member of research analyst's household.
Morgan Stanley Research policy is to update research reports as and when the Research Analyst and Research Management deem appropriate, based on developments with the issuer, the
sector, or the market that may have a material impact on the research views or opinions stated therein. In addition, certain Research publications are intended to be updated on a regular periodic
basis (weekly/monthly/quarterly/annual) and will ordinarily be updated with that frequency, unless the Research Analyst and Research Management determine that a different publication
schedule is appropriate based on current conditions.
Morgan Stanley is not acting as a municipal advisor and the opinions or views contained herein are not intended to be, and do not constitute, advice within the meaning of Section 975 of the
Dodd-Frank Wall Street Reform and Consumer Protection Act.
Morgan Stanley produces an equity research product called a "Tactical Idea." Views contained in a "Tactical Idea" on a particular stock may be contrary to the recommendations or views expressed
in research on the same stock. This may be the result of differing time horizons, methodologies, market events, or other factors. For all research available on a particular stock, please contact
your sales representative or go to Matrix at http://www.morganstanley.com/matrix.
Morgan Stanley Research is provided to our clients through our proprietary research portal on Matrix and also distributed electronically by Morgan Stanley to clients. Certain, but not all, Morgan
Stanley Research products are also made available to clients through third-party vendors or redistributed to clients through alternate electronic means as a convenience. For access to all
available Morgan Stanley Research, please contact your sales representative or go to Matrix at http://www.morganstanley.com/matrix.
Any access and/or use of Morgan Stanley Research is subject to Morgan Stanley's Terms of Use (http://www.morganstanley.com/terms.html). By accessing and/or using Morgan Stanley
Research, you are indicating that you have read and agree to be bound by our Terms of Use (http://www.morganstanley.com/terms.html). In addition you consent to Morgan Stanley processing
your personal data and using cookies in accordance with our Privacy Policy and our Global Cookies Policy (http://www.morganstanley.com/privacy_pledge.html), including for the purposes of
setting your preferences and to collect readership data so that we can deliver better and more personalized service and products to you. To find out more information about how Morgan Stanley
processes personal data, how we use cookies and how to reject cookies see our Privacy Policy and our Global Cookies Policy (http://www.morganstanley.com/privacy_pledge.html).
If you do not agree to our Terms of Use and/or if you do not wish to provide your consent to Morgan Stanley processing your personal data or using cookies please do not access our research.
Morgan Stanley Research does not provide individually tailored investment advice. Morgan Stanley Research has been prepared without regard to the circumstances and objectives of those

102
M
Foundation

who receive it. Morgan Stanley recommends that investors independently evaluate particular investments and strategies, and encourages investors to seek the advice of a financial adviser.
The appropriateness of an investment or strategy will depend on an investor's circumstances and objectives. The securities, instruments, or strategies discussed in Morgan Stanley Research
may not be suitable for all investors, and certain investors may not be eligible to purchase or participate in some or all of them. Morgan Stanley Research is not an offer to buy or sell or the
solicitation of an offer to buy or sell any security/instrument or to participate in any particular trading strategy. The value of and income from your investments may vary because of changes
in interest rates, foreign exchange rates, default rates, prepayment rates, securities/instruments prices, market indexes, operational or financial conditions of companies or other factors. There
may be time limitations on the exercise of options or other rights in securities/instruments transactions. Past performance is not necessarily a guide to future performance. Estimates of future
performance are based on assumptions that may not be realized. If provided, and unless otherwise stated, the closing price on the cover page is that of the primary exchange for the subject
company's securities/instruments.
The fixed income research analysts, strategists or economists principally responsible for the preparation of Morgan Stanley Research have received compensation based upon various factors,
including quality, accuracy and value of research, firm profitability or revenues (which include fixed income trading and capital markets profitability or revenues), client feedback and competitive
factors. Fixed Income Research analysts', strategists' or economists' compensation is not linked to investment banking or capital markets transactions performed by Morgan Stanley or the
profitability or revenues of particular trading desks.
The "Important Regulatory Disclosures on Subject Companies" section in Morgan Stanley Research lists all companies mentioned where Morgan Stanley owns 1% or more of a class of common
equity securities of the companies. For all other companies mentioned in Morgan Stanley Research, Morgan Stanley may have an investment of less than 1% in securities/instruments or
derivatives of securities/instruments of companies and may trade them in ways different from those discussed in Morgan Stanley Research. Employees of Morgan Stanley not involved in the
preparation of Morgan Stanley Research may have investments in securities/instruments or derivatives of securities/instruments of companies mentioned and may trade them in ways different
from those discussed in Morgan Stanley Research. Derivatives may be issued by Morgan Stanley or associated persons.
With the exception of information regarding Morgan Stanley, Morgan Stanley Research is based on public information. Morgan Stanley makes every effort to use reliable, comprehensive
information, but we make no representation that it is accurate or complete. We have no obligation to tell you when opinions or information in Morgan Stanley Research change apart from
when we intend to discontinue equity research coverage of a subject company. Facts and views presented in Morgan Stanley Research have not been reviewed by, and may not reflect information
known to, professionals in other Morgan Stanley business areas, including investment banking personnel.
Morgan Stanley Research personnel may participate in company events such as site visits and are generally prohibited from accepting payment by the company of associated expenses unless
pre-approved by authorized members of Research management.
Morgan Stanley may make investment decisions that are inconsistent with the recommendations or views in this report.
To our readers based in Taiwan or trading in Taiwan securities/instruments: Information on securities/instruments that trade in Taiwan is distributed by Morgan Stanley Taiwan Limited ("MSTL").
Such information is for your reference only. The reader should independently evaluate the investment risks and is solely responsible for their investment decisions. Morgan Stanley Research
may not be distributed to the public media or quoted or used by the public media without the express written consent of Morgan Stanley. Any non-customer reader within the scope of Article
7-1 of the Taiwan Stock Exchange Recommendation Regulations accessing and/or receiving Morgan Stanley Research is not permitted to provide Morgan Stanley Research to any third party
(including but not limited to related parties, affiliated companies and any other third parties) or engage in any activities regarding Morgan Stanley Research which may create or give the
appearance of creating a conflict of interest. Information on securities/instruments that do not trade in Taiwan is for informational purposes only and is not to be construed as a recommendation
or a solicitation to trade in such securities/instruments. MSTL may not execute transactions for clients in these securities/instruments.
Morgan Stanley is not incorporated under PRC law and the research in relation to this report is conducted outside the PRC. Morgan Stanley Research does not constitute an offer to sell or
the solicitation of an offer to buy any securities in the PRC. PRC investors shall have the relevant qualifications to invest in such securities and shall be responsible for obtaining all relevant
approvals, licenses, verifications and/or registrations from the relevant governmental authorities themselves. Neither this report nor any part of it is intended as, or shall constitute, provision
of any consultancy or advisory service of securities investment as defined under PRC law. Such information is provided for your reference only.
Morgan Stanley Research is disseminated in Brazil by Morgan Stanley C.T.V.M. S.A. located at Av. Brigadeiro Faria Lima, 3600, 6th floor, São Paulo - SP, Brazil; and is regulated by the Comissão
de Valores Mobiliários; in Mexico by Morgan Stanley México, Casa de Bolsa, S.A. de C.V which is regulated by Comision Nacional Bancaria y de Valores. Paseo de los Tamarindos 90, Torre 1,
Col. Bosques de las Lomas Floor 29, 05120 Mexico City; in Japan by Morgan Stanley MUFG Securities Co., Ltd. and, for Commodities related research reports only, Morgan Stanley Capital Group
Japan Co., Ltd; in Hong Kong by Morgan Stanley Asia Limited (which accepts responsibility for its contents) and by Morgan Stanley Bank Asia Limited; in Singapore by Morgan Stanley Asia
(Singapore) Pte. (Registration number 199206298Z) and/or Morgan Stanley Asia (Singapore) Securities Pte Ltd (Registration number 200008434H), regulated by the Monetary Authority of
Singapore (which accepts legal responsibility for its contents and should be contacted with respect to any matters arising from, or in connection with, Morgan Stanley Research) and by Morgan
Stanley Bank Asia Limited, Singapore Branch (Registration number T14FC0118J); in Australia to "wholesale clients" within the meaning of the Australian Corporations Act by Morgan Stanley
Australia Limited A.B.N. 67 003 734 576, holder of Australian financial services license No. 233742, which accepts responsibility for its contents; in Australia to "wholesale clients" and "retail
clients" within the meaning of the Australian Corporations Act by Morgan Stanley Wealth Management Australia Pty Ltd (A.B.N. 19 009 145 555, holder of Australian financial services license
No. 240813, which accepts responsibility for its contents; in Korea by Morgan Stanley & Co International plc, Seoul Branch; in India by Morgan Stanley India Company Private Limited having
Corporate Identification No (CIN) U22990MH1998PTC115305, regulated by the Securities and Exchange Board of India (“SEBI”) and holder of licenses as a Research Analyst (SEBI Registration
No. INH000001105); Stock Broker (SEBI Stock Broker Registration No. INZ000244438), Merchant Banker (SEBI Registration No. INM000011203), and depository participant with National
Securities Depository Limited (SEBI Registration No. IN-DP-NSDL-567-2021) having registered office at Altimus, Level 39 & 40, Pandurang Budhkar Marg, Worli, Mumbai 400018, India;
Telephone no. +91-22-61181000; Compliance Officer Details: Mr. Tejarshi Hardas, Tel. No.: +91-22-61181000 or Email: [email protected]; Grievance officer details: Mr. Tejarshi
Hardas, Tel. No.: +91-22-61181000 or Email: [email protected]. Morgan Stanley India Company Private Limited (MSICPL) may use AI tools in providing research services.
All recommendations contained herein are made by the duly qualified research analysts; in Canada by Morgan Stanley Canada Limited; in Germany and the European Economic Area where
required by Morgan Stanley Europe S.E., authorised and regulated by Bundesanstalt fuer Finanzdienstleistungsaufsicht (BaFin) under the reference number 149169; in the US by Morgan Stanley
& Co. LLC, which accepts responsibility for its contents. Morgan Stanley & Co. International plc, authorized by the Prudential Regulation Authority and regulated by the Financial Conduct
Authority and the Prudential Regulation Authority, disseminates in the UK research that it has prepared, and research which has been prepared by any of its affiliates, only to persons who (i)
are investment professionals falling within Article 19(5) of the Financial Services and Markets Act 2000 (Financial Promotion) Order 2005 (as amended, the “Order”); (ii) are persons who are
high net worth entities falling within Article 49(2)(a) to (d) of the Order; or (iii) are persons to whom an invitation or inducement to engage in investment activity (within the meaning of section
21 of the Financial Services and Markets Act 2000, as amended) may otherwise lawfully be communicated or caused to be communicated. RMB Morgan Stanley Proprietary Limited is a member
of the JSE Limited and A2X (Pty) Ltd. RMB Morgan Stanley Proprietary Limited is a joint venture owned equally by Morgan Stanley International Holdings Inc. and RMB Investment Advisory
(Proprietary) Limited, which is wholly owned by FirstRand Limited. The information in Morgan Stanley Research is being disseminated by Morgan Stanley Saudi Arabia, regulated by the Capital
Market Authority in the Kingdom of Saudi Arabia, and is directed at Sophisticated investors only.
The information in Morgan Stanley Research is being communicated by Morgan Stanley & Co. International plc (DIFC Branch), regulated by the Dubai Financial Services Authority (the DFSA)

Morgan Stanley Research 103

M
Foundation

or by Morgan Stanley & Co. International plc (ADGM Branch), regulated by the Financial Services Regulatory Authority Abu Dhabi (the FSRA), and is directed at Professional Clients only, as
defined by the DFSA or the FSRA, respectively. The financial products or financial services to which this research relates will only be made available to a customer who we are satisfied meets
the regulatory criteria of a Professional Client. A distribution of the different MS Research ratings or recommendations, in percentage terms for Investments in each sector covered, is available
upon request from your sales representative.
The information in Morgan Stanley Research is being communicated by Morgan Stanley & Co. International plc (QFC Branch), regulated by the Qatar Financial Centre Regulatory Authority
(the QFCRA), and is directed at business customers and market counterparties only and is not intended for Retail Customers as defined by the QFCRA.
As required by the Capital Markets Board of Turkey, investment information, comments and recommendations stated here, are not within the scope of investment advisory activity. Investment
advisory service is provided exclusively to persons based on their risk and income preferences by the authorized firms. Comments and recommendations stated here are general in nature. These
opinions may not fit to your financial status, risk and return preferences. For this reason, to make an investment decision by relying solely to this information stated here may not bring about
outcomes that fit your expectations.
The trademarks and service marks contained in Morgan Stanley Research are the property of their respective owners. Third-party data providers make no warranties or representations relating
to the accuracy, completeness, or timeliness of the data they provide and shall not have liability for any damages relating to such data. The Global Industry Classification Standard (GICS) was
developed by and is the exclusive property of MSCI and S&P.
Morgan Stanley Research, or any portion thereof may not be reprinted, sold or redistributed without the written consent of Morgan Stanley.
Indicators and trackers referenced in Morgan Stanley Research may not be used as, or treated as, a benchmark under Regulation EU 2016/1011, or any other similar framework.
The issuers and/or fixed income products recommended or discussed in certain fixed income research reports may not be continuously followed. Accordingly, investors should regard those
fixed income research reports as providing stand-alone analysis and should not expect continuing analysis or additional reports relating to such issuers and/or individual fixed income products.
Morgan Stanley may hold, from time to time, material financial and commercial interests regarding the company subject to the Research report.

Registration granted by SEBI and certification from the National Institute of Securities Markets (NISM) in no way guarantee performance
of the intermediary or provide any assurance of returns to investors. Investment in securities market are subject to market risks. Read
all the related documents carefully before investing.

INDUSTRY COVERAGE: Software

Company (Ticker) Rating (As Of) Price* (06/05/2025)

Chris Quintero
BILL Holdings Inc (BILL.N) O (01/16/2025) $45.68
Blackline Inc (BL.O) O (09/29/2024) $58.22
E2open Parent Holdings Inc (ETWO.N) E (01/17/2024) $3.23
OneStream Inc (OS.O) O (01/16/2025) $29.07
Vertex Inc. (VERX.O) O (01/17/2024) $41.87

Elizabeth Porter, CFA

Amplitude Inc. (AMPL.O) E (01/13/2025) $12.97
Autodesk (ADSK.O) O (08/23/2024) $298.22
Freshworks Inc (FRSH.O) E (10/18/2021) $15.61
GoDaddy Inc (GDDY.N) E (07/19/2021) $180.39
HubSpot, Inc. (HUBS.N) O (03/21/2023) $602.61
Klaviyo, Inc (KVYO.N) E (10/16/2023) $33.95
LegalZoom.com Inc (LZ.O) U (07/28/2022) $9.20
Liveramp Holdings Inc (RAMP.N) E (01/13/2025) $33.12
Semrush Holdings Inc -A (SEMR.N) O (01/13/2025) $10.11
Sprinklr Inc (CXM.N) E (07/19/2021) $8.85
Sprout Social Inc (SPT.O) E (11/17/2020) $21.86
Wix.Com Ltd (WIX.O) O (01/13/2025) $153.25
Zeta Global Holdings Corp (ZETA.N) E (08/01/2024) $13.64
ZoomInfo Technologies Inc (GTM.O) E (02/01/2024) $9.97

Josh Baer, CFA

Asana Inc (ASAN.N) U (05/20/2025) $14.72
BigCommerce Holdings, Inc. (BIGC.O) E (05/11/2021) $5.12
Box Inc (BOX.N) E (05/21/2024) $37.93
CCC Intelligent Solutions Holdings Inc (CCCS.O) O (11/13/2024) $8.92
Coursera, Inc. (COUR.N) E (06/05/2025) $8.86
DigitalOcean Holdings Inc (DOCN.N) O (01/16/2025) $27.77
Docebo Inc. (DCBO.O) E (05/12/2025) $26.85
DocuSign Inc (DOCU.O) E (01/16/2024) $92.90
Lightspeed POS Inc. (LSPD.N) E (02/18/2021) $11.34

104
M
Foundation

Sabre Corp (SABR.O) E (03/16/2021) $2.70

ServiceTitan Inc (TTAN.O) E (01/06/2025) $114.55
Toast, Inc. (TOST.N) O (12/16/2021) $42.82
Udemy Inc (UDMY.O) E (06/05/2025) $7.55

Keith Weiss, CFA

Adobe Inc. (ADBE.O) O (07/31/2023) $415.20
Akamai Technologies, Inc. (AKAM.O) E (04/29/2020) $76.14
Atlassian Corporation PLC (TEAM.O) O (01/13/2020) $218.60
Check Point Software Technologies Ltd. (CHKP.O) E (10/16/2023) $231.11
Cloudflare Inc (NET.N) O (12/02/2024) $176.33
CoreWeave (CRWV.O) E (04/22/2025) $135.05
CrowdStrike Holdings Inc (CRWD.O) O (01/09/2024) $462.94
CyberArk Software Ltd (CYBR.O) E (01/09/2024) $392.18
Fortinet Inc. (FTNT.O) O (10/07/2022) $103.50
Gen Digital Inc. (GEN.O) E (06/07/2024) $29.13
Intuit (INTU.O) O (02/26/2025) $766.64
Jamf Holding Corp (JAMF.O) E (10/13/2024) $10.54
Microsoft (MSFT.O) O (01/13/2016) $467.68
Okta, Inc. (OKTA.O) O (12/02/2024) $104.18
Oracle Corporation (ORCL.N) E (01/15/2019) $171.14
Palo Alto Networks Inc (PANW.O) O (10/10/2017) $197.11
Qualys Inc (QLYS.O) U (02/09/2021) $139.70
Rapid7 Inc (RPD.O) E (08/11/2015) $23.17
SailPoint Inc (SAIL.O) E (03/10/2025) $18.02
Salesforce, Inc. (CRM.N) O (12/21/2023) $267.14
Samsara Inc (IOT.N) E (03/23/2023) $47.25
SentinelOne, Inc. (S.N) E (12/02/2024) $18.12
ServiceNow Inc (NOW.N) E (10/21/2024) $1,017.60
Shopify Inc (SHOP.O) O (04/19/2024) $105.03
Snowflake Inc. (SNOW.N) E (02/29/2024) $210.10
Tenable Holdings Inc (TENB.O) E (12/02/2024) $32.56
Varonis Systems, Inc. (VRNS.O) O (10/16/2023) $50.14
Workday Inc (WDAY.O) E (02/19/2025) $250.10
Zscaler Inc (ZS.O) E (01/12/2023) $300.88

Meta A Marshall
8x8 Inc (EGHT.O) U (06/14/2024) $1.77
Five9 Inc (FIVN.O) E (10/10/2022) $28.55
NICE Ltd. (NICE.O) O (10/16/2023) $175.96
RingCentral Inc (RNG.N) E (08/08/2023) $26.80
Twilio Inc (TWLO.N) O (02/24/2025) $120.80
Zoom Video Communications Inc (ZM.O) E (10/11/2022) $81.33

Roy D Campbell
Karooooo Ltd (KARO.O) O (04/27/2021) $57.86

Sanjit K Singh
Appian Corp (APPN.O) E (12/03/2021) $31.80
C3.ai (AI.N) U (01/04/2021) $25.34
Confluent, Inc. (CFLT.O) E (01/16/2025) $24.18
Couchbase, Inc. (BASE.O) E (08/16/2021) $19.46
Datadog, Inc. (DDOG.O) E (01/16/2025) $121.75
Dynatrace Inc (DT.N) E (02/13/2024) $54.94
Elastic NV (ESTC.N) O (12/16/2024) $86.29
GitLab Inc (GTLB.O) O (10/09/2024) $48.77
JFrog Ltd. (FROG.O) O (12/21/2023) $43.18
MongoDB Inc (MDB.O) O (04/12/2023) $225.38

Morgan Stanley Research 105

M
Foundation

PagerDuty, Inc. (PD.N) E (01/24/2024) $15.38

Palantir Technologies Inc. (PLTR.O) E (02/04/2025) $119.91
UiPath Inc (PATH.N) E (09/07/2022) $13.26

Stock Ratings are subject to change. Please see latest research for each company.
* Historical prices are not split adjusted.

INDUSTRY COVERAGE: Internet

Company (Ticker) Rating (As Of) Price* (06/05/2025)

Brian Nowak, CFA

Airbnb Inc (ABNB.O) U (12/06/2022) $137.29
Alphabet Inc. (GOOGL.O) O (08/11/2015) $168.21
Amazon.com Inc (AMZN.O) O (04/24/2015) $207.91
Booking Holdings Inc (BKNG.O) E (01/09/2019) $5,570.52
DoorDash Inc (DASH.O) O (02/21/2024) $215.84
Expedia Inc. (EXPE.O) E (01/09/2019) $173.38
Instacart (CART.O) E (01/29/2024) $46.24
Lyft Inc (LYFT.O) E (10/24/2019) $15.53
Meta Platforms Inc (META.O) O (03/20/2023) $684.62
Nextdoor Holdings Inc (KIND.N) U (05/12/2025) $1.61
Pinterest Inc (PINS.N) E (03/28/2022) $33.76
Reddit Inc (RDDT.N) O (12/08/2024) $112.24
Snap Inc. (SNAP.N) E (07/22/2024) $8.20
Uber Technologies Inc (UBER.N) O (06/04/2019) $84.67

Matthew Cost
AppLovin Corp (APP.O) O (04/10/2025) $414.14
Compass, Inc. (COMP.N) E (07/20/2022) $6.16
Criteo SA (CRTO.O) E (01/26/2016) $26.28
DoubleVerify Holdings Inc (DV.N) E (06/25/2024) $15.03
Electronic Arts Inc (EA.O) E (08/04/2021) $147.88
Integral Ad Science Holding Corp. (IAS.O) E (04/16/2024) $8.24
Opendoor Technologies Inc (OPEN.O) E (07/24/2023) $0.66
Playtika Holding Corp (PLTK.O) E (11/27/2022) $4.88
Roblox Corporation (RBLX.N) O (11/04/2024) $94.20
Shutterstock Inc (SSTK.N) E (07/28/2022) $17.86
Take-Two Interactive Software (TTWO.O) O (02/01/2018) $231.03
Trade Desk Inc (TTD.O) O (06/01/2023) $71.11
Unity Software Inc (U.N) O (09/02/2024) $24.79
Webtoon Entertainment Inc (WBTN.O) E (07/22/2024) $8.78
Yelp Inc (YELP.N) U (01/10/2019) $36.80
Zillow Group Inc (Z.O) E (04/18/2018) $69.80

Nathan Feather
Bumble Inc. (BMBL.O) E (03/08/2021) $5.35
Chewy Inc (CHWY.N) O (10/31/2023) $47.49
Duolingo Inc (DUOL.O) O (04/23/2025) $522.99
eBay Inc (EBAY.O) O (04/18/2024) $77.74
Etsy Inc (ETSY.O) U (04/18/2024) $62.72
FIGS, Inc. (FIGS.N) E (02/29/2024) $5.09
Match Group Inc (MTCH.O) E (04/18/2024) $31.50
Peloton Interactive, Inc. (PTON.O) E (03/14/2022) $6.94
Revolve Group Inc (RVLV.N) E (10/20/2024) $22.57
WW International Inc (WGHTQ.PK) NR (05/20/2025) $0.27

Stock Ratings are subject to change. Please see latest research for each company.
* Historical prices are not split adjusted.

106
M
Foundation

© 2025 Morgan Stanley

Morgan Stanley Research 107