Julien Chaumond

Y Combinator CEO Garry Tan recently said that in about 25% of current YC startups, AI is writing 95% of the code. We’re in the endgame now — the marketing game. If building becomes faster and cheaper thanks to AI, the hardest part — and the key to success — will be how you position, brand and distribute. Here’s what I’m seeing: #1 Focus on niche markets: Go after $10M–$100M markets. Startups backed by traditional VCs usually skip these because they’re not aiming for a $1B market.But with AI, you can build quicker and leaner — which makes smaller markets not just viable, but strategic. #2 Build something that spreads on its own: You need a product with word-of-mouth energy — something that feels so cool, unique or personal people just share it.Think: lovable.dev, arcads.ai #3 Don’t stop at generic AI use cases: Support agent, email agent, sales agent — they’re good starting points, but crowded already. Use them as a playground, then go deeper. #4 Think ahead: what will AI agents need? When AI agents start running more and more workflows, they’ll need tools, data, services — their own ecosystem. Start building for them, not just for humans. (And no, not generic infra or monitoring — that’s already owned by the big players.) #5 Models will be commodities: So your edge? Marketing, UX, positioning, brand, niche. Just my 2 cents — what do you think?

65

19 Comments

Most founders are secretly miserable. But their LinkedIn feeds tell a different story: "Just hit $1M ARR!" "Featured in TechCrunch!" "Thrilled to announce our Series A!" Meanwhile, they're: • Living on ramen • Haven't slept properly in months • One bad quarter away from shutting down I've seen founders sacrifice everything to maintain the appearance of success. They chase VC money because it validates their existence in Silicon Valley. They chase unicorn status so others view them as the next Zuck. Meanwhile, most successful entrepreneurs I know run "boring" businesses. They solve real problems, generate actual revenue, and sleep well at night. They don't need a TechCrunch headline to feel accomplished. Lesson there.

278

99 Comments

Agents & LLMs : YC is calling it out loud and clear... 1) Full-Stack AI – Don’t just sell software to incumbents - replace them entirely with AI-native companies. 2) Designer Founders – Design is a competitive edge. It’s time for designers to build, not just beautify. 3) Voice AI – Phone calls haven’t changed in 100 years. Voice bots are now indistinguishable from humans. 4) AI for Science – Bring AI into drug discovery, chemistry, and materials science to reshape physical industries. 5) AI Personal Assistants – Move from "to-do" lists to “done” lists. Let LLMs handle your tasks. 6) Healthcare AI – $1T+ wasted on admin. AI can fix broken workflows and connect siloed systems. 7) AI Tutors – Every learner deserves a personal tutor. AI can now deliver learning in an interactive, multimodal, and smart way. 8) Robotics Software – Robotics is on the verge of its ChatGPT moment. Build tools to make it real. 9) Future of Education – Reinvent how 1.5B students learn. AI can personalize and scale instruction like never before. 10) AI Home Security – Ring was just the beginning. AI can truly protect and personalize residential security. 11) Internal Agent Builders – Every employee will soon build their own AI agent. Who’s building the infrastructure? 12) AI Research Labs – It’s not just about quick products. YC backs deep, long-horizon AI research too. 13) Voice AI for Email – Your voice. Your inbox. On the go. Build assistants that actually do the work. 14) AI for Personal Finance – Democratize smart, unbiased financial advice. Zero-cost, personalized, and private. To look into this in a different way... 🧠 AI-First Infrastructure & Tools - Full-Stack AI Companies – Rebuild entire businesses - Internal Agent Builders – every employee automate their own work - AI Research Labs – Deep-tech R&D with long-term impact - Robotics Software Tools – next generation of intelligent robots 🎨 Creative & Product-Led Innovation - Designer Founders – Empower product designers to be founders - AI Personal Assistants – Assistants that learn your routines - Voice AI for Email – Use your voice to triage and respond 🏥 Healthcare & Science - Healthcare AI – Automate administrative bloat in the $4T U.S. healthcare system - AI for Scientific Advancement – Speed up research in chemistry, biology, and materials science 📚 Education & Learning - AI Tutors for Everyone – Personalized, multimodal learning for any topic - Future of Education – Reinvent classroom learning using LLMs 📞 Communication & Customer Experience - Voice AI – Replace outdated call centers with human-like voice agents 🏠 Consumer & Residential Tech - AI Home Security – Smart, AI-driven protection 💰 Finance & Wealth Management - AI for Personal Finance – Provide affordable, intelligent, unbiased financial advice at scale. Building with LLMs or agents? there’s never been a better time. https://lnkd.in/e9vCVrGf #startups #founders #entrepreneurs #YC #accelerators #application #AI

63

6 Comments

Romain Huet at OpenAI DevDay demo‘ing an idea that I think will be HUGE. Like super huge. I think this will be pervasive in 6-12 months: Users of your app will have a new “Codex” tab that will allow you to modify the software and add new features in-place. Inside the app itself. No code editor, no PRs, no deployment. This demo was for a weight loss management app. But I think we will have a lot of enterprise / internal software add in-place app edits. They’ll gate it to admin users.

305

44 Comments

I recently joined the OVHcloud podcast to unpack how the venture capital landscape is evolving—and what it means for founders building in today’s market. 💥 Key takeaways: - VCs are going earlier than ever → Pre-seed is now institutionalized. Angels have evolved from check writers to strategic allies. - AI is reshaping capital flows → Foundation models are consolidating, but there’s still room to win in the application and infrastructure layers. - Europe’s B+ funding gap is real → US capital is still essential post-Series B. VC should help founders prepare for this from day one. - Open Source AI → OSS faces challenges (compute, data, regulation), but remains a powerful force in infrastructure, tooling, and domain-specific models. Thanks to OVHcloud and Juliette Otterburn Hall for the great conversation and continued partnership (from Dataiku to Quandela) 🎙️ Listen to the episode: https://lnkd.in/ez75rjrM 📊 Our Commercial Open Source report: https://lnkd.in/ei4pqDGS Serena

45

2 Comments

We sold Stilt in 2022, but that’s not what created lasting value. Here’s the true value created: 15% of our team becoming founders. Great startups create a virtuous cycle: Find exceptional talent → they become founders → they build successful companies → the ecosystem grows stronger Take our alumni network: - Multiple got accepted into Y Combinator - $25M+ collectively raised - Success across multiple geographies (US, Canada, India) eg Karthik, one of our early engineers, launched Scale3 and Langtrace. He raised $5.3 million from Redpoint. At Stilt, we understood that team-building is the core responsibility of the founders. We optimized for the best talent possible: - People with founder potential - Those who could outgrow us - Individuals we'd work for if roles were reversed And surrounded them in the right environment: - Autonomy to develop leadership skills - Cross-functional exposure - Founder mentality cultivation As Vinod Khosla says: "The team you build is the company you build." Most founders focus on product-market fit. But the truly transformative companies? They don't just build products — they build founder factories. Startups that generate exponential impact beyond their original vision are the true legends.

67

6 Comments

Very interesting news: Mixpanel just launched Heatmaps 🔥 + Session Replay for web — and now mobile 📱. This essentially eliminates Amplitude's edge. But it also raises a bigger question: do tools like Fullstory, Heap, and Hotjar still have a place? Let’s be honest — the main reason many teams added these tools was for visual insights into user behavior. If that’s now available inside Mixpanel (which I’ve always seen as the strongest tool for event-based analytics), why keep paying for another stack of tools? 👉 Is this the beginning of the end for Hotjar / Heap / Fullstory? Or do they still bring unique value? #ProductAnalytics #Mixpanel #ProductManagement #UXAnalytics The Product Alliance

51

4 Comments

When I was in YC, I was surprised to see that startups are now vibe-coding their entire companies in 3 days. Not weeks. Not months. Days. — Here’s the exact $40 stack making it possible: - Lovable/v0 by Vercel → frontend in hours - Cursor/Claude Code → AI pair programmer - Supabase → backend in a box - Stripe → collect payments - Willowvoice.com → Prompt AI with voice, makes coding feel like magic — How the process works: Day 1 — Ship the core logic Start with the simplest version of your product. Forget design or polish. Just get the main feature working (even if it’s just a single button). Day 2 — Talk it into existence Instead of making wireframes in Figma, founders literally talk through what happens with WillowVoice to an AI coding tool: “User clicks this → then sees that…” That narration gets transcribed and fed into AI tools to generate components, layouts, and dashboards. What used to take a day can now be done in an hour. Day 3 — Tighten the flow Once the product works, the final step is smoothing out the onboarding and user flows. Instead of typing long prompts, founders just talk through their fixes with WillowVoice, which AI implements quickly. — What YC founders are learning: 1. Perfect code is the enemy of shipped code 2. AI works best when iteration speed is insane 3. The first version will be wrong anyway → build it in days, not months — The result: A working product in 2–3 days. Crude? Sure. But enough to launch, test, and iterate. YC startups are proving it: the future of development is AI + vibe-coding at the speed of thought. Would you build your next idea this way, or do you think speed comes at too high a cost???

1,207

129 Comments

Just had a conversation with a group of venture investors. They have an in-house team testing LLMs — not with benchmarks, but with prompts designed to replicate 100 real products they backed since 2020. Not futuristic “super agents”, but innovative tools with a real-world complexity score of 5 to 8 out of 10. Today’s top models can replicate just 2% of those products without code. With GPT-5, they’re expecting 22–25%. If that happens, they believe the venture landscape — as we know it — won’t survive. I’m sharing my SMB-focused prompt systems with them. Right now, they automate ~30% of the daily operational workload for small businesses — but this is still in the "paper experiment" phase. I’m observing 20 boutiques, restaurants, and salons across 4 countries, tracking how they apply AI to real workflows. This isn’t the internet, where feedback loops take a week. Here, it takes months. But the early signs are promising. We’re nervously excited to test it all on GPT-5.

66

2 Comments

Fascinating to see how much interest and energy there is around enterprise AI right now — especially when the conversation moves beyond hype to the real-world challenges of building and scaling production-grade solutions. Yesterday at the OVHcloud x YourStory Media AI Startup event, my co-founder SK A.ya and I had the privilege of sharing our perspective on why enterprise AI code generation is still an unsolved problem — and how we’re tackling it at Toystack AI. The questions and insights from the audience reinforced just how relevant these topics are: everyone’s excited about the possibilities of AI, but equally aware of the gaps that need to be solved before it can truly deliver for enterprises at scale. The challenge: Enterprise-grade AI code often lacks structure, security, context-awareness, guardrails, and seamless cloud deployment. The approach we’re taking: At Toystack AI, we’ve built a system that combines multi-agent orchestration, a mix of LLMs, context-rich databases, best-in-class coding patterns, and strict guardrails — all fully integrated with automated, one-click Kubernetes cloud deployments. We also spoke about why a truly enterprise-ready code-gen solution can’t be built in isolation — it has to address real-world client needs and production constraints, not just idealized developer workflows. On the topic of scaling, I shared why getting to $100M+ revenue in this space will require more than the classic product-led growth motion — it calls for a sales-led, custom-service approach that resonates with enterprise realities. It was equally inspiring to learn from other founders doing amazing work in this space: Abhilash Salimath of TrueFan AI, enabling creators to monetize directly and deliver personal wishes — even from Rishabh Pant, in any language! Arvind Venugopal Nair of NeuroPixel.AI, building AI-powered, hyper-personalized marketing and creative intelligence — imagine any clothes on any model, in any setting. Really enjoyed the thoughtful discussion on trends in AI and valuation of AI startups, moderated by Shivani Muthanna of YourStory Media, with insights from a stellar panel: Shubham Jhuria (Aeravti Ventures) Nishant Chandra (Newton School) Ankit K. (Capital-A) Sriram Kannan (Routematic) A big thank you to Satyam Santosh for organizing and to OVHcloud for showcasing the powerful compute and cloud capabilities they’re bringing to India and beyond. The energy in the room was incredible — and the engagement on these tough-but-important topics reminded me how crucial it is to keep building with both ambition and real world pragmatism. #AI #EnterpriseAI #CodeGeneration #LLM #AgenticAI #Toystack #Cloud #DevOps #OVHcloud #AIStartups #Valuation #YourStory

68

3 Comments

🚨 The battle for the future of software is no longer about who can write code fastest — it’s about who owns the data and context to power AI. Generative coding agents and tools from Anysphere, GitHub Copilot, Windsurf Cognition and poolside are commoditizing software development. In this new era, context is king and data is the moat. Winners will be those who: 1️⃣ Embed deeply into workflows to capture rich, high-fidelity data 2️⃣ Engineer context precisely at inference time 3️⃣ Reinforce models through usage, not static prompts Legacy SaaS vendors face eroding moats. AI-native apps with data flywheels and contextual intelligence are rising fast. 👇 Full article in the comments. If you’re building in this space, I want to hear from you. Thank you to Jeff Haynie, Jason Warner, Çağla Kaymaz and Vedant Agrawal for providing input and feedback on this article! #AI #Software #VC #DeveloperTools #ContextIsKing #FutureOfWork

64

10 Comments