Gil Akos

Today, we’re thrilled to publicly launch Phoebe and announce $17M in seed funding, in rounds led by GV and Cherry Ventures. We are building the immune system for software: An AI agent platform that continuously investigates live data, diagnoses emerging issues and generates preemptive fixes. As AI accelerates how software is built, reliability must keep pace. Phoebe started with faster troubleshooting, and our aim is to prevent the overwhelming majority of emergent problems before becoming customer-impacting failures. Huge thanks to the hundreds of early users at customers such as Trainline and PPRO. Thanks to our investors, including 20 CTOs and tech veterans joining as angels. And of course, thanks to the small and mighty Phoebe team! This is just the beginning! 🚀 Shoutout to Mark Kleinman at Sky News for breaking the news this morning (linked in the comments) Some of the many backers we're grateful for 🤗 🙏 GV (Google Ventures), Cherry Ventures, Anthony Danon, Carmen Alfonso Rico 🍫, William Gaybrick, Shola Akinlade, Yaron Shaer, Christian Dorffer, Tom Stafford, Thomas Wolf, Andrew Tan, Victor Riparbelli, Harsh Sinha, Jeanne DeWitt Grosser and more

682

140 Comments

🚀 Big news for AI agent builders: Agent Reinforcement Trainer (ART) now integrates seamlessly with 🦜🔗 LangGraph! If you’ve ever built multi-step, ReAct-style agents, you know the pain: endless prompt tweaking, fragile reasoning, and lots of manual trial-and-error. With this new integration, you can now train your LangGraph agents with reinforcement learning — making them smarter, more adaptive, and more efficient over time. ✨ What this means in practice: * Agents learn to reason through multi-step problems without hand-holding * They improve tool usage automatically (no more prompt spaghetti) * They make smarter decisions in complex workflows * Training is scalable and robust, powered by ART’s reinforcement engine And the best part? It’s a drop-in upgrade: * Swap in ART’s init_chat_model and wrap_rollout * All interactions, tool calls, and reasoning chains get logged automatically * Plug straight into ART’s RULER reward function for training — no custom rewards needed 💡 Imagine your LangGraph prototypes evolving into production-ready agents that actually get better with use. That’s the power of combining LangGraph’s agent framework with ART’s reinforcement learning. 👉 Full docs & examples here: https://lnkd.in/gwWj-xxx Can’t wait to see what the community builds with this. If you’re experimenting with multi-step AI agents, give your agents the ability to learn from experience!

1,887

80 Comments

The single biggest determinant of your career success is how early you can be exposed to excellence. Graham Duncan shared this insight on his podcast with Patrick, and it has stuck with me since selling my company to Ramp as their first acquisition when they were just 80 employees. That experience fundamentally shaped how I operate and think about building companies, and I wanted to share three lessons I've carried forward to NationGraph: 1. Don't wait for permission to take action. This is a core value that Eden Ding and I have at NationGraph. This isn't "don't wait for someone to tell you how to do your job" it's a mentality shift. Don't wait for permission to dominate an industry. Think about expense software: traditional players like Expensify or SAP Coupa are making air travel beautiful within airport constraints—champagne as soon as you arrive, fast pass through the security line, straight to the nicest lounge, lie-flat seats on the aircraft, premium shuttles to the hotel. But this is not the best airport experience. The truly great airport experience is how Ramp approached the problem, like a private jet. Walk up to a simple terminal, no champagne, hand your bags to someone, get on the plane, and go. No one wants the smoothest way to file expense reports; they don't want to file expense reports at all. Ramp approaches all problems in this first principles way anchored on saving companies time and money. We do the same at NationGraph. 2. Metal sharpens metal Sharp debate, hard work, and absurdly high standards bring out the best of teams, but balance all that with genuine kindness. I've learned so much from Geoff but this really stuck with me. Someone on his team wasn’t scaling with the organization fast enough, and instead of letting them go, Geoff focused on finding a role where they would succeed and push the company forward. This person was intense and exceptional at their core strengths, and that kindness really stuck with me. You brought them onto your team for a reason. Your team has more to offer than just the role you brought them on for; always be looking for where they can power the organization forward. 3. Create momentum in everything you do. Winning truly is contagious. When people feel they’re on a winning team, it creates this sense of inevitability to accomplishing the mission that becomes self-fulfilling. The best companies don’t just execute well—they create environments where excellence becomes inevitable. Congratulations and great work to my friends and old teammates from Ramp, it's a joy to watch you all succeed. :Ramp-intensfies: https://lnkd.in/ezjbvWVi

70

1 Comment

🚀🚀 AI is rewriting the rules of startup success. In our latest Crunchbase piece, Noah Lin and I dive into how we @ defy.vc have seen AI-first companies scale smarter with value model tactics—leveraging predictive power and network effects to reshape industries. From defying traditional metrics to building data moats, these strategies are a game-changer for founders. 💡 Stay tuned for a deeper follow-up where we’ll share actionable tactics for founders to scale their AI ventures! Full article 👇 #AI #Startups #Innovation #VentureCapital

50

4 Comments

We teamed up with The Product Folks on mosAIc 2025 - a virtual build a thon for AI builders. As part of the buildathon, there are several hands-on workshop that covers everything from evals, prompting, and agent design to security, infra, and shipping production-ready flows. This is what it actually takes to build a working agentic application. Alongside speakers such as Arvind, Ji Huang, Madhu, Akash, Rohit Khanna, Rahul, Vaibhavi, Eashan, I'll be doing a hands-on workshop as well, to answer questions like: → What truly goes into building an AI agent that can do real meaningful job - layers involved in building it → How do agents work securely on behalf of users and connect with external tools to get the job done. → How should builders think about auth where agents are the primary actors If you’re working on AI infra or products, this is probably the most useful 2 weeks you’ll spend this year. Last few spots left. Link in the comments.

70

5 Comments

An agentic workflow startup called Sim Studio (YC X25) spent $220,000 to acquire the Sim.ai domain name from Andy Booth about a week ago. Sim had been using SimStudio.ai for its website. https://lnkd.in/e32c9y89

52

4 Comments

Today, we are releasing our next milestone in building general superintelligence at Deep Cogito. We are releasing 4 hybrid reasoning models (70B, 109B MoE, 405B, 671B MoE) as a proof of concept for a novel AI paradigm - iterative self-improvement (AI systems improving themselves). Highlights: 📌  We are releasing 4 hybrid reasoning models of sizes 70B, 109B MoE, 405B, 671B MoE under open license. 📌  The largest 671B MoE model is amongst the strongest open models in the world. It matches/exceeds the performance of the latest DeepSeek v3 and DeepSeek R1 models both, and approaches closed frontier models like o3 and Claude 4 Opus. 📌  We extend our work on building superintelligence using Iterated Distillation and Amplification (IDA) by scaling the model’s intelligence prior. This is done by the model internalizing the reasoning process using iterative policy improvement, rather than simply searching longer at inference time. This is in line with the approach used for AlphaGo / AlphaZero, applied to natural language. 📌 This is a novel scaling paradigm where the models develop more “intuition”, and serves as a strong proof of concept for self-improvement. Since the Cogito models develop a better intuition of the trajectory to take while searching at inference time, they have 60% shorter reasoning chains than Deepseek R1. We are excited to get our research and models into the hands of the AI community. You can download the models on Hugging Face, or use them as an API on Together AI or Runpod or Baseten, or run them locally using Unsloth AI. We expect significant improvements by self improvement over the next few months using our approach. All models we create will be open sourced. More details in the blog below.

352

44 Comments