The agent ecosystem is evolving rapidly and we are trying to capture a picture of where we are and where things are headed. We're now collecting responses for the 2026 survey of agents. Help us create a more complete picture of how agent systems are evolving, what challenges remain, and where future research may be most impactful.

New in @berkeley_ai blog. Cao et al. create new kernel optimizer K-search and use it to auto adapt CUDA to MLX. Their attention translation measures 0.97x speed to Apple’s native implementation. Translate your code with their OSS tool. Available now. 🐻📄

KV cache management is arguably one of the hardest problems in LLM inference, and advancing it requires the community to learn from and build on one another’s work. Great to see that happening. Amazing work from @lightseekorg on adopting and evolving vLLM's hybrid memory allocator! The Jenga paper from @ChenZha62999224 is especially relevant today as attention architectures become increasingly complex.

"We're spending about five figures a month on coding agents. I think it's making the team more productive. I'm not sure. But I think." That's from a VP of engineering running a team of ~100. It's obvious that coding agents are faster than human engineers, but it's not obvious how you manage your spend. For decades, the constraint on shipping software was headcount. You had N engineers, you pointed them at the highest-priority work, and you shipped what you shipped. Today you can spend 2-3x the cost to produce the same software in a tenth of the time. The recent port of Bun from Zig to Rust took 11 days and $165k in API costs. That's obviously a net improvement over what it would have cost to have engineers do it by hand, but choosing to do that requires a new calculus. Because faster production is not the same as better economics. The fact that you can generate more code, faster, doesn't mean you should. Every task is now a choice: which model, how much speed, how much accuracy, how much human guidance. Get those parameters wrong and you pay a lot of money to build the wrong thing very fast. The challenge for engineering teams has always been picking the right things to build. Agents didn't fix that problem — it's faster but just as expensive:

Best Paper Award to our ICML Workshop paper 'Fantastic Adaptive Taxonomies and How to Use Them'. Let me summarize briefly what the paper is about: Failure taxonomies are becoming increasingly important. We show that failure taxonomies can be used in multiple ways: 1. As a test-time scaling tool for best-of-N judges, 2) as a mutation feedback mechanism in optimization loops and 3) as runtime feedback for coding agents. Previously people used MAST and other hand-made fixed failure taxonomies. In this paper we show how to create the failure taxonomy adaptively and dynamically: We observe agent rollouts and create an adaptive failure taxonomy, bespoke to the agent weaknesses and task challenges. These adaptive taxonomies give a massive boost: On Terminal Bench 2 we get 89.9% with Opus 4.6 / Forgecode harness and the adaptive taxonomy used with a Best-of-N Judge, outperforming fixed taxonomies by 15%. (1/n)

Intelligence sufficient for knowledge work is rapidly becoming free. What does this mean for the future of data systems and agents? A new blog post outlines a new research agenda @Berkeley_EECS; 🧵

⚙️ Speaker Spotlight: Joseph Gonzalez Every time you use an LLM — or check which model tops the leaderboard — there's a good chance his research made it possible. @profjoeyg is a professor at @UCBerkeley and co-director of the Sky Computing Lab, whose group built some of the most-used open-source AI systems of the LLM era. 🔹 Professor of Computer Science at @UCBerkeley · co-director, Sky Computing Lab 🔹 His group created vLLM, Chatbot Arena (now LMArena), Gorilla & Vicuna 🔹 Co-founder of RunLLM · previously co-founded Turi/GraphLab (acquired by Apple) 🔹 A leading voice on LLM agents, serving & how we evaluate AI While others build the models, he builds the systems that run them — and the arenas that rank them. Hear Joseph on stage at AGI Summit SF 2026. 📅 July 18–19, 2026 📍 Palace of Fine Arts, San Francisco 🎟 Tickets: 🏷️ 15% off with code GenAI-26 #agisummit# #aiareall# #AIsystems#

Yesterday at Open Frontier, organized by @LaudeInstitute in San Francisco, it felt like we were at the beginning of something historic. Not a conference. A working meeting. A room full of people who will help define the future of open AI.

What if a robot policy weren't a neural net or a test-time chat loop, but a multi-file code repo selected from a Pareto frontier of genetically evolved candidates? RHO moves all its LLM exploration to training time, then runs that repo on scenes it was never trained on. 🧵👇🏽

❤️ vLLM is PagedAttention, but quickly expanded to the platform for open source models, frontier hardwares, and efficient inference for all workload at speed of light! Today, the PA kernel is no longer used as all kernels support it natively; the scheduler and engine remained.

LEANN / @YichuanM, @andylizf, @sewon__min, @matei_zaharia, @profjoeyg (@UCBerkeley, @Princeton) A low-storage RAG system that enables fast, accurate, and fully private retrieval directly on personal devices.

...and thanks again to @matei_zaharia @istoica05 @lateinteraction @dawnsongtweets @gneubig @pgasawa @JonSaadFalcon @heathercmiller @ryan_marten @alexgshaw @profjoeyg @swyx and Ioannis Ioannidis!

Two years ago, @profjoeyg and I wrote that open models wouldn't catch the frontier — but that it didn't matter. That's more true than ever today. The whole debate is framed around the wrong question. Everyone keeps asking whether open models will beat GPT or Claude on evaluations. But almost nobody building real applications needs the single best model in the world for every task. They need a model that's smart enough, fast enough, and cheap enough — running where they can count on it. That's a much lower bar, and open models cleared it a while ago. As access to frontier models gets more complicated for application builders, "close to frontier and fully under my control" starts to look like the better deal, not the consolation prize. GLM 5.2 landing in Cursor is the kind of signal that makes this concrete — a model you can actually reach for, not a benchmark you admire from a distance. The frontier labs will keep selling the best top-end model. That was never the contest worth winning for most of us. Smart enough, fast enough, cheap enough is good enough:

Built by @YichuanM and @andylizf at StarTrail, out of UC Berkeley's Sky Computing Lab @BerkeleySky, BAIR @berkeley_ai, and Berkeley NLP @BerkeleyNLP. Same team behind LEANN (RAG on everything, on-device, with a 97% smaller index). Thanks to our advisors @matei_zaharia, @profjoeyg, and @sewon__min! If LEANN gave your models memory of your private data, PixelRAG gives them eyes on the open web. Code: Paper: Playground: 🐍 𝚙𝚒𝚙 𝚒𝚗𝚜𝚝𝚊𝚕𝚕 𝚙𝚒𝚡𝚎𝚕𝚛𝚊𝚐

At @CAISconf last month, @andykonwinski sat down with researchers on the conference floor -- @matei_zaharia @istoica05 @lateinteraction @dawnsongtweets @gneubig @pgasawa @JonSaadFalcon @heathercmiller @ryan_marten @alexgshaw @profjoeyg @swyx and Ioannis Ioannidis -- to talk open vs. closed, agent evals, compound AI systems, and where the frontier is headed 👀

Try our new image browsing plugin for Claude Code, built by the legend @andylizf 🚀 It started as the ultra-efficient screenshot crawling pipeline we built for PixelRAG. Turns out screenshot-based browsing is often more token-efficient, more accurate, and preserves layouts/tables/images better than text-only browsing. Now you can plug it directly into Claude Code with one command. Code: Install: uv tool install pixelrag Give it a try and let us know what you build! 🔥

Something I haven't advertised much is that I made a Discord to go with my RLHF book, launching in print in a few weeks. Trying to create the place for the next generation of folks trying to learn post-training to learn and have community.