Aldrick ZAPPELLINI

McKinsey just released the “State of AI 2025” Report ↳ They reveal how businesses are adopting AI ↳ What separates AI winners from the rest ↳ The challenges businesses are facing TL;DR: → AI adoption surges (78% overall, 71% use generative AI). → Big companies lead, restructuring workflows and hiring AI talent. → CEO oversight boosts impact, especially in large firms. → Few track KPIs, slowing AI value capture. → AI reshapes jobs, driving reskilling over layoffs. → Revenue up in sales, supply chain; costs down in HR, IT. → Service jobs shrink, tech roles grow. → Success needs full AI transformation, not experiments. Key Takeaways: → AI adoption is rapidly growing, but most companies are still early in capturing value. → Large companies (> $500M revenue) are adopting AI faster and restructuring workflows. → CEO involvement in AI governance strongly correlates with financial impact. → Less than one-third of organizations follow best practices like tracking KPIs. → AI talent demand remains high, but hiring difficulty is easing slightly. → More employees are being reskilled rather than replaced. → AI is boosting revenue in marketing, sales, and supply chain. → Cost reductions are most visible in HR, risk & compliance, and IT. → Jobs in customer service and supply chain are shrinking, while IT and software engineering roles are growing. → Companies that fully transform their business with AI (rather than experimenting) will win. AI isn’t just a tool. It’s a business transformation strategy. Is your business adapting fast enough?

265

125 commentaires

#OpenSource is just like open science: shared, collaborative, and the foundation of progress. 🚀 At PyData Paris 2025, I had the privilege of moderating the panel “Open-source Business” with Yann Lechelle (CEO, :probabl., former CEO of Scaleway) and Sylvain Corlay (CEO, QuantStack) at the Cité des sciences et de l'industrie. Both leaders are deeply embedded in projects like scikit-learn and Project Jupyter—the invisible infrastructure used daily by millions from scientists to business analysts. Our conversation unpacked what it really takes to grow sustainable businesses on top of open source.: • No one-size-fits-all: Every project needs its own model; hybrids are the norm. • Community = momentum. Every user is an advocate. Marketing spend shrinks when you invest in shared value • Safeguard the team: Visibility comes with risks; support and protection are non-negotiable. • Diversify funding. From VC to contracts to public investment (like France’s support of scikit-learn), sustainability depends on many levers. • Play the long game. The most critical projects are often the least glamorous. Openness as strategy. Beyond code: open science, data, source, weights, standards, hardware. A European playbook for digital sovereignty—with France and Germany catalyzing all 27 EU countries 🇪🇺 🗽. Sustainable open-source businesses balance community values with commercial viability—and “openness” is not ideology; it’s an industrial strategy that lets society understand and control its destiny. Thanks to Yann and Sylvain for a candid, deeply practical discussion. Looking forward to Yann's book "ouvertarisme.fr" that talks about how openness (from open science to open source to open models) can become an industrial policy. In fact, it has become one. We need to seize the occasion without falling into cynicism nor "open washing”.

126

3 commentaires

As some of you may have noticed, I'm explaining the architecture, roadmap and vision of the Actian Data Intelligence Platform in a series that I simply call Actian Explained. The roadmap and overall vision is crafted by one of the most brilliant tech pioneers in France, Guillaume Bodet. Guillaume introduced the Scrum methodology in France, in partnership with Jeff Sutherland, and as such he is one of the earliest adopters of modern software engineering globally. I met Guillaume as he was transitioning from software into data - but he didn't transition like the majority of software engineers did. Because he also stayed in software engineering, as he is building software - then Zeenea, now the Actian Data Intelligence Platform with many new features and capabilities, well beyond a regular data catalog. It's the coming together of modern software principles applied in the data domain, which are so sorely needed, _directly_ in a software product intended for managing data at scale. When I talk about our data intelligence platform, I change from explaining already existing components in the platform that are ingeniously crafted, over the release of innovative features, to our long term visions and distinct point of view on how companies should work with data. So far, in this series, I have explained Episode 1: The Data Intelligence Platform: https://lnkd.in/dKhUU-CM Episode 2: Data Observability: https://lnkd.in/dixsMUBM Episode 3: Data Products and Data Contracts: https://lnkd.in/dP3TMCvv Episode 4: Federated Knowledge Graph: https://lnkd.in/dw5tHQ66 Episode 5: Model Context Protocol for AI applications: https://lnkd.in/dfsDNgD5 Episode 6: Ask AI: Natural Language Search https://lnkd.in/dAZUn7xx Episode 7: Metadata Management https://lnkd.in/dP5j68nU Episode 8: Data Management as Business Process https://lnkd.in/d7DQ28cD #data #dataintelligence #dataobservability #datacatalog #datamanagement #dataengineering #datascience #datagovernance #dataproduct #datacontract #accessmanagement #datalineage #dataarchitecture #datamodeling #datadiscovery #dataanalytics #scalabledatamanagement #datalake #datalakehouse #datawarehouse #metadata #metadatamanagement

73

4 commentaires

🚨 McKinsey & Company's latest report on the ‘AI Bank of the Future’ just dropped — with some powerful updates. What’s new? 🔹 A more complete GenAI architecture From tools ➜ to agents From analytics ➜ to orchestration From point solutions ➜ to full-stack, GenAI-native architecture At the core of this transformation is multigent orchestration, where specialized AI agents are coordinated by a central AI “brain” that plans, routes tasks, and triggers the right agent at the right time. 🔹 **From “use cases” to domain rewiring Instead of building scattered pilots, McKinsey suggests focusing on a few high-impact subdomains (like credit, onboarding, or fraud) and transforming them end-to-end. 🔹 A stronger emphasis on reuse and scale Reusable AI components Shared orchestration logic AI control towers to drive governance and coordination across teams. In short: The vision has expanded, but the path is now clearer and more detailed. While it’s still early for most banks — and probably most industries — to adopt such a stack, concepts like agents, orchestration, and domain-level transformation are starting to feel more universally relevant. I share my journey here. Join me and let's grow together. Aditya Kachave #AI #GenAI #McKinsey #AIArchitecture #OpenSource #AgenticAI #AIOrchestration #AdityaKachave #Innovation

32

7 commentaires

GPT-5 is 🤯🤯 People trashing it don't know how to use it. We created a guide for my portfolio companies. An overview: 1. Encourage Deeper Reasoning ⤷ Explicitly instruct GPT-5 to “think harder” 2. Use Explicit Planning Phases ⤷ Ask it to turn requests to components, identify ambiguity, create a plan & validate 3. Be Highly Explicit in Instructions ⤷ Define tone, style, structure & output expectations 4. Structure Prompts Rigorously ⤷ Organize information into defined sections 5. Request Thought Process Summaries ⤷ Ask it to explain its reasoning before answering 6. Avoid Conflicting Instructions ⤷ Identify and resolve contradictions in your prompt 7. Leverage Self-Rating and Iteration ⤷ Tell it to rate its answer and iterate till it gets a 10/10 answer 8. Apply Metaprompting ⤷ Ask it to improve your prompt 9. Adjust Verbosity ⤷ Define how long its answer should be Our full guide has details, prompt examples, and actual use cases. Comment “GPT” to get it.

63

168 commentaires

For those of you thinking about execution impact models this week, a recent paper worth reading is 𝙂𝙚𝙣𝙚𝙧𝙖𝙩𝙞𝙣𝙜 𝙍𝙚𝙖𝙡𝙞𝙨𝙩𝙞𝙘 𝙈𝙚𝙩𝙖𝙤𝙧𝙙𝙚𝙧𝙨 𝙛𝙧𝙤𝙢 𝙋𝙪𝙗𝙡𝙞𝙘 𝘿𝙖𝙩𝙖 by Guillaume Maitrier, Gregoire Loeper & jean-philippe bouchaud. It tackles a well-known challenge: most impact studies rely on proprietary datasets—costly, limited in scope, and often biased. This paper shows how to recreate the key stylised facts of metaorders—Square Root Law (SQL), concave execution profile, and post-trade decay—using only public trade data and a synthetic metaorder generation algorithm. In my opinion, it’s practical, cleanly implemented, and replicable. What I find particularly interesting is that the synthetic metaorders behave indistinguishably from real ones, even though there's no embedded alpha or signal. That underscores something most execution models still underplay: the mechanical nature of price impact. If synthetic flow without intent leaves the same footprint, then the marginal cost of liquidity may be more about flow mechanics than trader belief. This raises a bigger question: if we can replicate impact dynamics purely from flow properties and order book behavior, how much of what we model as “cost” is just structural? Mega job done by the authors for making this approach both transparent and reproducible — a rare and valuable contribution for quants on both the research and implementation side, thank you! 🙌 Link: https://lnkd.in/dzakb3PY

61

6 commentaires

"DeepSeek shows that betting on a single LLM provider will be a losing game… the ability of new players to disrupt the landscape in a single weekend makes it clear: companies need an LLM-agnostic approach." — Florian Douetteau, Dataiku co-founder and CEO As our CEO Florian shares in InfoWorld, DeepSeek’s innovative mix of multi-token prediction, precision training, and reasoning reward models shows how fast AI is evolving. Open-access breakthroughs like these push the field forward, but enterprises must stay flexible, maintain control, and build AI architectures that can adapt. 🔗 Read more: https://bit.ly/4kPllcZ | #AI #LLM #DeepSeek

73

1 commentaire

𝗧𝗵𝗲 𝗚𝗲𝗻 𝗔𝗜 𝗥𝗲𝗮𝗹𝗶𝘁𝘆 𝗖𝗵𝗲𝗰𝗸 𝗡𝗼 𝗢𝗻𝗲 𝗪𝗮𝗻𝘁𝘀 𝘁𝗼 𝗛𝗲𝗮𝗿 (𝘍𝘳𝘰𝘮 𝘢 𝘸𝘦𝘣/𝘮𝘰𝘣𝘪𝘭𝘦 𝘷𝘦𝘵 𝘸𝘩𝘰 𝘫𝘶𝘮𝘱𝘦𝘥 𝘪𝘯 2 𝘺𝘦𝘢𝘳𝘴 𝘢𝘨𝘰 - 𝘣𝘶𝘪𝘭𝘵 3 𝘈𝘐 𝘱𝘳𝘰𝘥𝘶𝘤𝘵𝘴, 𝘤𝘰𝘯𝘴𝘶𝘭𝘵𝘦𝘥 𝘧𝘰𝘳 15+ 𝘦𝘯𝘵𝘦𝘳𝘱𝘳𝘪𝘴𝘦𝘴) After 20+ years in web/mobile and 2 years deep in AI implementations, the data paints a sobering picture: The numbers don't lie: → 𝗗𝗲𝗹𝗼𝗶𝘁𝘁𝗲 2024: Only 30% of Gen AI experiments are expected to scale → 𝗠𝗰𝗞𝗶𝗻𝘀𝗲𝘆 2023: 92% of firms are increasing AI investments, but just 1% consider themselves "mature" → Avg. enterprise Gen AI spend ~$1.9M, yet <30% of executives report clear ROI The technical reality I'm seeing: 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗹𝗶𝗺𝗶𝘁𝘀: Many models still max out at 32K tokens (despite newer 128K options) 𝗠𝗲𝗺𝗼𝗿𝘆 𝗴𝗮𝗽𝘀: No true session retention without costly workarounds   𝗛𝗮𝗹𝗹𝘂𝗰𝗶𝗻𝗮𝘁𝗶𝗼𝗻𝘀: Error rates of 15-20% in complex business tasks 𝗔𝗰𝗰𝘂𝗿𝗮𝗰𝘆 𝘀𝘁𝗿𝘂𝗴𝗴𝗹𝗲𝘀: LLMs falter in nuanced decision-making Gartner placed Gen AI in the "Trough of Disillusionment"—and I see why. Where each tech actually wins: ✅ Gen AI: Content creation, marketing, customer service ✅ Traditional ML: Fraud detection, predictions (anything needing 99%+ accuracy) Even Marc Andreessen admits models are "hitting a capability ceiling." My take: We're forcing problems into AI hype rather than matching tech to real needs. The cost of Gen AI dropped 1000x in 2 years—but that doesn't make it the right hammer for every nail. 💡 Agree? Disagree? Share your AI scaling struggles below. 🔄 Repost to help others avoid the hype trap. #AI #MachineLearning #TechLeadership #AIReality #EngineeringStrategy

19

3 commentaires

🌍 BREAKING: The UN just made history on AI governance. For the first time, the United Nations General Assembly has adopted a resolution creating two global mechanisms to guide how humanity navigates artificial intelligence: 🔹 Independent International Scientific Panel on AI – 40 experts worldwide, tasked with producing evidence-based assessments on AI’s risks, opportunities, and impacts. 🔹 Global Dialogue on AI Governance – an annual forum for governments, industry, civil society, and academia to build shared approaches and bridge divides. 💡 Why it matters: AI is advancing faster than policy. Without coordination, we risk fragmented rules, growing inequalities, and weakened trust. These mechanisms aim to: • Promote safe, secure, and trustworthy AI • Support capacity-building in developing countries • Ensure transparency, accountability, and human oversight • Advance open data, open models, and interoperability ⚖️ For now, this resolution is non-binding — but it lays the foundation for international cooperation. Much like other global compacts, its true power lies in shaping norms, aligning priorities, and building trust across borders. This is a historic step: AI governance is moving from national debates to the global stage. 👉 What do you think — will global cooperation help ensure AI serves humanity, not the other way around? Link to the full resolution in comments 👇 #AIforGood #AIGovernance #GlobalGoals #UN

162

58 commentaires