Olearis

Anthropic files for IPO by end of August. OpenAI lines up for September. So half of the AI products in the world are about to run on the APIs of two publicly traded, quarterly-earnings-reporting companies. And the question nobody wants to say out loud: how exactly does this affect the rest of us who just want to keep shipping? Let us actually break it down. 𝐖𝐡𝐚𝐭 𝐢𝐬 𝐜𝐨𝐨𝐥 𝐚𝐛𝐨𝐮𝐭 𝐢𝐭? More money in the ecosystem, more stability signals, easier enterprise sales when your vendor is a public company (procurement loves that), a real chance the platforms invest harder in reliability, SLAs, and enterprise-grade tooling. Long term, that is good for anyone building on top. 𝐖𝐡𝐞𝐫𝐞 𝐚𝐫𝐞 𝐭𝐡𝐞 𝐮𝐧𝐝𝐞𝐫𝐰𝐚𝐭𝐞𝐫 𝐫𝐨𝐜𝐤𝐬? Public companies plan quarter by quarter. Which means model deprecations get faster (old models are a margin drag), pricing gets squeezed the moment growth cools (hi, everyone who remembers "unlimited plans"), and roadmaps quietly start bending toward whatever Wall Street rewards next. The "we love developers" energy meets the "what did we tell the market?" energy, and one of those wins. 𝐖𝐡𝐚𝐭 𝐬𝐡𝐨𝐮𝐥𝐝 𝐲𝐨𝐮 𝐝𝐨 𝐭𝐡𝐢𝐬 𝐰𝐞𝐞𝐤? Nothing dramatic. But it is a good moment to ask yourself: if my API bill doubles tomorrow, do I still have a business? If my model gets deprecated on 90 days notice, do I have a fallback ready? Where does an open-weight option (Kimi, Qwen, DeepSeek, Llama) fit in my stack as insurance? questions for the CTOs and founders reading this: Single provider or already hedging with multi-model routing? Anyone accelerating an open-weight fallback specifically because of the IPO news? Or is this all just noise and you are staying the course? Drop your real take, not the LinkedIn take 👇

We are getting ready to drop a fresh batch of case studies on the Olearis website, and there are a couple we are honestly excited to talk about. 🍫 One of them is an AI-powered hospitality platform we built for a chocolatier who is a bit of a household name in Singapore and Sydney. Think recipes, suppliers, food costs, all in one place, with AI doing the heavy lifting for chefs who would much rather be creating than fighting spreadsheets. 🎯And if you happen to enjoy throwing darts (or just watching other people do it), stay tuned. We shipped a really cool one in that space too, with on-device AI scoring, online matchmaking, the works. That story is coming soon Full case studies landing on our site shortly. Follow along if you want a look behind the scenes at what we have been building lately 👀

Have you heard about Kimi K3? Of course you have, the internet has been buzzing about it for the last three weeks. The thing everyone keeps testing it on is penetration testing, because Claude politely refuses to help with that kind of thing and Kimi just goes ahead and does it. Open weights, free to self-host, and it reasons through attack surfaces like a red-teamer who had one coffee too many. ❗️Now the awkward part. It's a Chinese model, you're doing security work, on systems that really shouldn't be leaking anywhere, especially not somewhere with a reputation for enthusiastic data collection. Yes, you can self-host it (that's the whole point of open weights), but every time a client's infrastructure ends up in the prompt window, there's still that tiny voice in your head asking if this is a great idea. So, question for the security folks who are already playing with it: Are you using Kimi K3 for pentesting? Self-hosted only, or hitting the API? How do you handle the "great model, spicy origin" thing? Really curious how the community is thinking about this one 

Can your dating app tell a real selfie from an AI face? If not, here's how you'll find out: a user gets scammed, posts about it, the story spreads, and your app-store rating is in freefall by Friday. The cleanup takes a year, but the prevention takes a week. Until apps catch up, here's how to check a match yourself (fair warning: a determined scammer can fake every one of these now: treat them as flags, not proof): • Reverse-image the photos - drop them into Google Lens or TinEye (best for faces). If they show up nowhere else, or trace to a stranger's Instagram, that's your answer. • Zoom in on the edges - ears, teeth, hands, jewelry, background text. Generators still smear the details they bet you won't check • Ask for a live pose, not just a video call - "turn your head sideways" or "hold up a word on paper." Real-time deepfakes glitch or dodge; a real person does it in two seconds. (Plain video calls stopped being proof in 2026.) • Rushing you to Telegram or WhatsApp? 🚩 flag... moving off-app early is how scammers escape moderation. every one of these is a manual patch on a problem the app should solve. that week of building trust in (verification, stolen-photo detection, real-time deepfake and love-bomb checks) 𝐢𝐬 𝐞𝐱𝐚𝐜𝐭𝐥𝐲 𝐰𝐡𝐚𝐭 𝐰𝐞 𝐝𝐨!

"My app is 95% done, I vibe-coded it myself, just need someone to ship it to the App Store" 📱 We get this message every week. Every. Week. And we love it, why? You built the thing, that's the hardest part. But then we open the project and… well. That "last 5%" tends to be where the entire iceberg is hiding: — API keys hardcoded straight into the app — works great with 3 test users, melts down with 1,000 — Apple rejects it on day one for a privacy string nobody mentioned — payments "work" until someone actually tries to get a refund — change one thing → three other things break Our CTO Oleg wrote the honest version, what AI coding tools are genuinely great at, what they quietly skip, and how to tell if your "almost done" app needs a rescue or just a polish. Read it before you ship 👇 https://lnkd.in/d6EDibsG P.S. If you've built something with AI and got stuck in that last stretch - send it over. We do honest audits of vibe-coded apps and AI-built MVPs, rescue the broken parts, and ship products that actually survive real users. App Store compliance, security, scalable architecture, payments, the whole boring-but-critical layer. 📩 hello@olearis.com — vibe-coded app rescue, MVP audit, mobile app development, iOS, Android, Flutter, AI product development.

Last week, Olearis turned 13 🎂 13 years of product releases, office moves, full remote shifts, meetings on opposite sides of the planet, raft trips, late-night pizza, and the kind of team chat banter that turns into our own cultural code (guys we love you!) But mostly - 13 years of people. A special shout-out to those who've been with Olearis since day one (yes, we really do have colleagues like that, and we're incredibly proud of it). And to every teammate who's joined the journey since - you're the reason Olearis is Olearis. And a huge thank you to our clients - the products we've shipped together over these years are the real measure of what Olearis is. Trusting us with your ideas, your roadmaps, and sometimes your wildest "is this even possible?" moments - that's the partnership we're most grateful for. Here's to year 14 💚

Friday, 6pm. 437 applications in the inbox. Your recruiter is crying into her third coffee 🥺☕️ Spoiler: 380 of those resumes were written by ChatGPT. The one person you actually wanted got auto-rejected at minute one because she had a gap year. And somewhere in your ATS, a rule from 2019 that nobody remembers writing just killed your next senior hire. Welcome to hiring in 2026 - where the bots apply, the bots screen, and the humans are exhausted. We wrote about how to actually fix this: → https://lnkd.in/dXA9gZtG

In 2026, the bottleneck in telehealth is clinician attention. And AI triage is how the best platforms are spending it wisely. We just published a new piece on AI telemedicine app development - specifically, how intelligent patient triage and prioritization are quietly becoming table stakes for any telehealth platform that wants to still be around in 2028. Inside the article: • What an AI triage layer actually does (intake, acuity scoring, specialty routing, ambient documentation) • Market numbers worth putting in your next board deck • Custom platform vs. off-the-shelf SaaS — when each makes sense • A practical tech stack for HIPAA-compliant telehealth • FDA pathways, costs, timelines If you're building or scaling a telehealth product (or your clinicians are drowning in documentation) this one's for you. 🔗 https://lnkd.in/dcwj3f8R #Telemedicine #HealthcareAI #DigitalHealth #AIinHealthcare #HIPAA #HealthTech #Olearis

Half of this year's AI-driven layoffs are already being walked back. Three real cases worth looking at: - Klarna replaced around 700 customer service jobs with AI, watched quality drop on complex cases, and is now rehiring. CEO Sebastian Siemiatkowski's exact words: "we went too far." - IBM automated 94% of HR with its AskHR agent — but kept humans for the remaining 6% (sensitive cases, ethical calls, anything that needs empathy). Total headcount actually went up. The savings got redirected, not pocketed. - Duolingo's "AI-first" memo got dragged on social media so hard the CEO publicly walked it back and removed AI usage from performance reviews. The data agrees. Forrester (Predictions 2026): 55% of employers regret the AI cuts they made. Gartner: half will be reversed by 2027, often under a different job title. The question before any AI rollout isn't "can it replace this team?" It's "what does it do on the weird stuff?" That's where every one of these projects breaks. Anyone seen this play out at their own company?

Last weekend, a robot ran a half marathon faster than any human ever has. Honor's Lightning finished 21 km in 50:26 (nearly 7 minutes ahead of the world record) fully autonomously, using tech adapted from smartphones and self-driving cars. A year ago, robots at the same race could barely finish. Now they're breaking records. That's what happens when you stop reinventing the wheel and start connecting existing expertise to new problems. And honestly, it's never been a more exciting time to do it.