Today we're announcing AdaL. Our goal: true autonomy — agents and humans moving as one, starting with developer and coding agents.

Coding agents can run nonstop in the cloud or create a PR all on their own, yet we are still babysitting them. Why? Because when a coding agent starts working in a repo, it generates alien code: duplicate functionality, overengineering, and anti-pattern design. Human engineers with (1) a global picture of the codebase and (2) an understanding of how the architecture and features are evolving are therefore required to review the code. Otherwise, you get “codebase rot”: bloated code that is impossible to understand, debug, or build upon. The next generation of agents will have human-level autonomy. For coding, that means getting more than 90% of PRs done autonomously without a single human engineer in the loop. That is where team productivity increases by another 10x.

think of watching an agent sending tasks to ur coding agent in real-time, be it @adalagent or claude code. it works fully autonomously but just like ur tesla, if it got into an emergency, u can instantly take over. how would you react? on this upcoming pre-launch party at @Soma_Capital office in SF, u will know. we let guests try AdaL Engineer come to hang with the team and other serious builders 👉 luma.com/o5an8a43

come to explore the next gen agent! luma.com/o5an8a43

Next Wednesday we’re heading over to Soma Capital for a co-working night with some of the best agent builders in SF in collaboration with @buildclub_. Secure your spot, link in comments :)

What model are you defaulting to right now and why?

Every Wednesday we put our agents on YOLO mode and walk to the farmers market. Burritos cherries walnuts. Work hard play hard I guess?

big launch soon!

as a founder, you can never feel as content as the moment when ur vision matures into clarity when all past struggles accumulate into something extraordinary that doesn’t resemble anything in the world and when you see a future full of promise and hope here is to the next chapter of @adalagent

RT @panda_liyin: finally, we can do h1b🚀🚀🚀 we are hiring come to build the best harness infra and memory model to power the world's agents

We are hosting a small prelaunch party with @Soma_Capital 🎉where builders will get early access to @adalagent Grab a drink 🥂, bring your laptop, and build alongside engineers working on coding agents, browser agents, and more. RSVP below 🔽

Getting an agent to execute is solved. Getting one you can trust with real work is not. That was the conversation all day at the AI-tonomy Summit, where our founder @panda_liyin showed AdaL live to a room of builders, researchers, and founders at @PlugandPlayTC Thanks for having us!

as ai took over the code generation, it creates a new gap: understanding @adalagent self-review is to close this gap and to help engineers (1) keep all code changes in one tab, instead of splitting across local and the github pr (2) cluster changes and walk though engineers to understand and improve self-review is invented to stop the vibe coding crisis.

gonna do a prelaunch party at @Soma_Capital SF on the night of 17th June. builders will get the early access to @adalagent agent sdk, meta-agent, and adal engineer which likes humans, operates a coding agent to work. if ur in sf, come to hang: luma.com/o5an8a43

Instantly switch between specialized agents for deep research, browser use, coding, self-review and sub-agent delegation with just /agent. Along with /model, it unifies your entire workflow, so you always have the right model and toolset active without breaking your flow state to open another tab.

Manual prompts are reactive. /cron makes them proactive. With /cron, AdaL can check builds every 5 minutes, detect new failures, and return a concise triage summary automatically. Your scheduled prompt goes through the exact same pipeline as manual input, so @ file references, tool usage, and multi-step reasoning all work identically. Set it and forget it.

How does AdaL stack up against Claude Code? We put the numbers up publicly. On coding (SWE-Pro-Curated-50), AdaL solved 50% of tasks vs Claude Code’s 46%, and in 34% less time, with 60% fewer tokens, and at $0.68 per success vs $1.87. On browser automation, we hit 92% on BU100. That’s 2.5 points above the next best agent. Fewer tool calls, fewer tokens, and better results is what our unified agentic workflow system gets you. Learn more here: adalagent.ai/benchmarks

a coding agent will not be truly autonomous w/o browser use. last night, @adalagent in browser use mode autonomously fixed a long-backlogged issue, all while i was watching a movie with my friend @AnnieLiao_2000 . its like the great engineer i always wished i had.

Yesterday, we hosted Agents After Dark with Build Club, and it was incredible getting to meet so many talented builders making things here in SF. Thanks to everyone who came out and made it such a great evening. See you at the next one! Sign up here: luma.com/o5an8a43

try it out: adalagent.ai/landing/opus-4-…