In academia we learn math and algorithms that teach us how to best model the world, but we rarely learn how to build a production grade harness that can handle 200 QPS without race conditions lol. Building agentic systems is a software engineering problem as much as a Research one. We need to start treating our training environments with the same rigor we treat production -- seeing enough RL envs get thrown out on this is why I wrote this series :)

We just raised a $60M Series B at a $1B valuation to build @daytonaio into the Agent Cloud. @FirstMarkCap led the round, with all key existing investors and angels doubling down. We're also thrilled to welcome the founders of some generational companies aboard, including @patrickc (Stripe), @tobi (Shopify), @AravSrinivas (Perplexity), @BrendanFoody (Mercor), @oliveur (Datadog), @adamdangelo (Quora) and @ericsyuan (Zoom) We're a 35-person team. But that's about to change; we're hiring across the board, and we'd love for you to join us.

You guys aren’t ready for October 27

not enough. would love to see a race to the bottom on training prices

Metas Muse Glimmer tech report is pretty interesting. Metas Muse Glimmer report describes training agents inside simulated digital lives, complete with 10–14 contacts, weeks of messages, calendars and documents. The agents practice tasks spanning multiple days, while training trajectories containing hallucinated results or unverified claims of success are discarded.

We looked at TerminalBench 4.0 and found a lot of issues. After fixing the issues Sonnet 5.5 scores 89.2%. So, TerminalBench is already mostly saturated. Along the way we found a lot of crazy stuff, including reward hacks in the published transcripts.

dressing down in sf is a form of respect towards the other person, its like bowing in japan, the lower you go the more respect you have

a lil something to take the edge off

Other honeypot ideas to entice your researcher lover: offer to buy and pay for a Waymo, and just have them dropped off at a secure location where you can extract the information efficiently! As long as there's a Waymo and some MK snacks, this will probably be enough. Hope this helps. 🤗

Exicteddddd I also hear the wonderful @mcshanimal will be joining us 👀

One more thing we have not yet released, and it's coming later this month – a tech report. We'd like to share all the cool science that went into Beam. But we want to do this write-up justice, so it's going to cook for a bit longer. More to come!

monthly proof of life

high agency low ego is my favorite personality combination. s-tier to work with in the age of intelligence.

made banana bread gelato this weekend! it is truly my best work yet banana gelato with layers of banana bread pieces

Today, we're launching Sazabi version 1.0. The reality of shipping software as a team in the age of AI is actually crazy. Agents merging their own PRs. Hundreds of deployments per day. A codebase that is constantly changing. Junior developers with 10x or 100x leverage. 😵‍💫 I spent the last year building a safety harness for teams that ship with AI. It's called @sazabi. The premise is simple: 1. You send us logs. 2. We detect anomalies, investigate them, and send you back rich, actionable alerts about real problems in your app. Sazabi helps you find issues before your users notice. No monitors or noisy alerts. Just pure signal. And today we're making it available for anyone to use.

Honestly engaging with these two on their on has been some of the highlights of my time in the Bay Area so excited to see some of their hot takes make it to the internet 🤭 Congrats @SOTAwives !!

I am so proud of Beam. Beam is Reflection's first open-weight model. It's the result of a ~year of hard work, ingenuity, and camaraderie of one of the most incredible teams I've ever seen in AI 🔥 I joined Reflection last November. Pretraining in particular was so nascent – only 5-10 people and a bit of code! We didn't yet have a crawler, or a resilient GPU infra, or a deduplication pipeline, or in-house evals, or even sufficiently big MoEs. All that started in earnest in 2026, in one of the most satisfying sprints I've ever had the pleasure to live through. What we nailed, I think, was a perfect blend of scientific rigor, startup intensity, team iteration, and ambition. We did not YOLO decisions in pretraining, that tends to blow up in your face when you scale up 🙂 But we also made sure to move fast and play to our strengths as a nimble, high-ownership, high-trust team. The resulting system of systematic iterations and fast cross-team back-and-forths really paid off. The midtraining, RL, and post-training teams really really cooked here on top of the pretrained base. The announcement below details the sheer scale and payoff of that investment, I'll let the team cover it, it's remarkable. On many agentic coding capabilities the model kept going up without any sign of a plateau. Feel free to read more in the announcement today – but later this month we will also drop the full weights release under Apache 2.0, a detailed tech report with the cool LLM science, and partner integrations with the OSS ecosystem. It's an open model after all. I strongly believe open intelligence is the future of AI, this is why I'm here. And I encourage y'all to join us on the open side 😉

camp hard launch 2026 🩵

📸 a well spent weekend in the redwoods @hardlaunching @BainCapVC @Accel

Today we are open-sourcing 🥕Karotte, our framework for building RL environments. We've used it for the past year to build MLE RL environments for frontier labs, and it's been hardened through 1M+ evaluation runs and red-teaming.