Enrique Herreros

We’ve been jamming with Y Combinator’s Summer 25 batch, and have been blown away by the quality and pace of the teams. What jumps out: -Infra > interface: some excellent teams are building in the execution fabric, trust rails, or evaluation loops. They’re tackling latency budgets, anti-bot arms-races and SOC-2 checklists: the plumbing that lets agents graduate from demos to enterprise workloads. -Data moats by design: the strongest vertical plays lead with proprietary claims corpora or HIPAA-grade call graphs. -Hard-tech is back: Flywheel’s tele-ops excavators and Verne's one-hour-training vision stack prove YC is funding robots again, but with a software-first margin profile. -Founders who’ve lived the pain: Ex-Cleartax PMs fixing GST reconciliation. Ex-Goldman post-trade leads building zero-trust for agents. Wet-lab scientists automating plate handling. Why we’re excited: -These teams aren’t asking whether agents will work; they’re solving the reliability, audit, and compliance gaps so agents can be boringly dependable. -Several are already in production with paying customers pre-Demo Day. -Defensibility comes from data moats, regulatory grounding and real infrastrucutre. You can see the agentic web moving from prototype to platform. Looking forward to seeing some of you tomorrow at Demo Day.

213

10 comentarios

Breaking: Sam Altman is no longer CEO of ChatGPT. Sam is still CEO of OpenAI overall. But Fidji Simo (ex-Instacart CEO) has stepped in as CEO of Applications now running ChatGPT, product, and operations. he is moving to mainly to long-term research, compute infrastructure, and capital planning. Even the best founders eventually face a choice: keep running operations or focus on growth work on the company > work in the company Structure isn’t weakness, it’s power. he's handing off his day-to-day, so he can free himself to work on stuff that no one else really can

106

16 comentarios

Lovable just raised $200M at a $1.8B valuation. We spent 48 hours in Stockholm with the team and saw how startups will be built in 2025: 1/ WHEN IT WORKS, IT PULLS Their story is insane. They launched GPT Engineer. It exploded 🤯 50k GitHub stars in days! Lovable was born to build the UI on top. Same thing happened with Replit. It took them 8 years to reach $10M ARR. Then, they found “vibe coding”. They dropped everything else. And hit $100M ARR in under a year. In AI, PMF doesn’t creep in. It hits like a wave. 2/ MARKETS WITH 2 LEADERS MATTER Two leaders = duopoly = closed market. CRM: Salesforce + Hubspot. eCommerce: Shopify + WooCommerce Lovable is betting on becoming the 1st player. That’s where the door is. 3/ TIME IS RELATIVE Everyone’s working weekends. Not out of pressure, but alignment. When everyone’s pulling in the same direction, speed compounds. 1 year at Lovable = 10 years in SaaS. 4/ YOUNG TALENTS I used to believe experience was everything... Turns out, some of the most brilliant builders hit their stride in their early 20s. Cursor’s founders graduated in 2022. Lovable’s average age looked like 24 😅 I felt like a dinosaur. 5/ RULES ARE MEANT TO BE BROKEN I asked Fabian, co-founder & CTO, who their persona was. He said: “anyone who wants to build an app.” At first, I didn't get it. At June, we narrowed down to find PMF. You don't always need to niche down in AI. Lovable is replacing Webflow, Framer, and Bubble... at the same time. 6/ GTM IS OVERRATED WHEN PULL IS STRONG Lovable hired Elena Verna. One of the best growth leaders out there. But traction came before. Lovable go-to-market is its community. People use it, share it, pay for it. GTM is here to accelerate things. It puts fuel on the fire 🔥 7/ THE OPPORTUNITY ISN’T OBVIOUS “Vibe coding” was dismissed because of retention concerns. But I’ve seen the data, and trust me, it’s wrong. This opportunity wasn't obvious to most people. And it helped them. 8/ FUNDRAISING IS SECONDARY This raise didn’t happen on a roadshow!! It came from a trusted conversation. Focus on building, and the right money finds you. 9/ COST ISN’T A PROBLEM In SaaS, AWS gave you free credits. In AI, LLM providers do the same 😉 If your product is great, infra will follow. 10/ THINK FROM SCRATCH The"code on the left, app on the right" UI was their invention. I asked Fabian how he came up with it. He said “it felt like the best UI”. Great products invent their own interface. Others copy it later. (Tinder for X, Uber for Y) ______ These 48 hours were a glimpse into what the next wave of startups might look like. Thanks again to Anton and Fabian for having us. For being so open, and the cold swim 🥶 You're building a generational European company 🇪🇺 And to Elena for making the intro. You're one of the greatest 💜

527

34 comentarios

This company might be the fastest-growing enterprise AI company in the world. In January, they were doing $1B in ARR. Now? Over $4B... in just six months. And they're not done. The Financial Times reported that Anthropic is raising a new round at a $150B+ valuation - more than double their last round. Even more surprising? They're considering sovereign wealth funds from the Middle East — a reversal from their earlier stance of rejecting such money on ethical grounds. CEO Dario Amodei reportedly told employees: “No bad person should ever benefit from our success” …but admitted that’s a hard principle to scale a company on. Why does this matter? Because in the AGI race, principles, talent, and capital are all colliding - in real-time. Anthropic has: - $8B+ committed from Amazon - A strategic partnership with Google - 80%+ of revenue from business customers - Claude, which is gaining serious ground in coding and RAG workflows But it’s still an uphill battle. OpenAI leads on consumer usage with 500M+ weekly ChatGPT users. Meta is handing out $100M researcher offers like candy. xAI is shipping fast with Musk’s distribution firehose. And yet - if Anthropic pulls this off, it won’t just be a valuation story. It’ll be a case study in how to build a $100B+ AI company without playing by Silicon Valley’s old rules. This is the quiet giant that might end up rewriting the map.

111

45 comentarios

LLMs alone aren’t enough for production environments. Sure, they can hit 70–80% accuracy. That’s impressive for demos. But in customer support, it’s not enough. When a customer is waiting, you need 95%+ accuracy: ✅ following policies, ✅ matching tone of voice, ✅ and resolving the issue end-to-end. That’s why we don’t just pitch “an AI agent.” We deliver a managed service,  with a promise to cut CS costs by 70%  without hurting CSAT. How? With an Agentic Architecture: 👉 Workflows and decision trees, just like your best agents follow 👉 Deterministic rules layered with LLMs 👉 Automation that hits 95% accuracy This is what it takes to run AI in production. Great thanks to Eytan Levit for the thoughtful interview 🙏 P.S. For the full interview check out Mixtiles's YouTube.

58

6 comentarios

Only ~3% of startups get into YC. Even fewer make it onto their Top 100 AI list. A while ago, StackAI did. 18 months later: - 100K+ AI agents deployed - enterprises in finance, healthcare, and defense - a team of 30+ from MIT, Meta, NASA, BCG How did we get there? By: - talking to users daily = fast feedback loops (early signals make the long-term vision more convincing) - building ugly (seriously, don't worry much) things fast - hiring slow, shipping fast, the right team multiplies velocity - staying uncomfortable - the moment it feels stable, you’re stagnating If you’re trying to get into YC, here's my 2 cents: - show real traction (even if it’s small), not hype - show technical depth, not noise - build a product that solves real pain, fast, scalable, and grounded P.S. StackAI is just getting started. We’re hiring, shipping weekly, and working with some of the most ambitious teams deploying AI into real operations. If you’re building something, or even just thinking about applying to YC, do it. You don’t need it all figured out. Just enough momentum, a real problem, and the courage to hit submit. Good luck! 🍀

185

31 comentarios

Other software are apparently already the primary AI users. Not people. Who is actually consuming the tokens? ✅ Google: 1.3 quadrillion tokens per month across its products ✅ OpenAI: 260T per month, 800M weekly users, 4M devs ✅ Groq: 50T+ per month (crazy amount of tokens and most ChatGPT users have never even heard of them) Do the napkin math: even with heavy chat use, humans cannot explain those totals. The gap is agents, batch jobs, RAG pipelines, evals, indexing, synthetic data. Invisible workloads at massive scale. Google is still so undervalued. They process 5x the amount of tokens compared to OpenAI, but very few talk about it. They have the deepest stack for builders: chips, infra, models, data, distribution and it shows up in tokens, but not headlines. Wall St analysts have been sleeping on Google, and still are. Chat is for demo. Machine to machine is where the money will be made.

64

13 comentarios