introducing Managed Agents on the Gemini API - in one API call, you get agent that comes with a remote Linux environment hosted by Google, ready to scale - you can define custom instructions, skills, and tools in Markdown

every major AI company now ships their own managed agent runtime and the pattern is converging fast, one API call, hosted environment, custom tools in config files, the race is now about whose execution layer you trust with your workloads

Lets go! 🚀

Expect insane bills from Google unless you buy the $200/mo Ultra plan

How am I gonna learn all of these bruh!!! @GoogleAIStudio pls tell me u got some tutorials or courses for learning all these

Managing agents is one thing, but wrapping them in a hosted remote Linux environment is the architectural shift we've been waiting for. This moves the bottleneck from 'how to orchestrate' to 'what to build.' Scaling agentic workflows just became a lot more predictable.

Theres no enough time to learn how Google is going to destroy AI battle! Plus quantum computing, full ecosystem, we have just started!

When could we use all this goodness with subscription OAuth instead of API?

One call = remote Linux env + custom Markdown tools/skills. 💪Who’s shipping with this first? 📷#GeminiAPI #BuildInPublic #AgenticAI

Is this the Gemini API or antigravity API now?

multi-region:eu possible?

The important part here is “one API call.” Google is trying to reduce the complexity of building autonomous AI systems dramatically.

Someone pls tell me how or what can I use these agents for 🙄

one agent in a sandbox is the solved case. the unsolved case: three managed agents that need shared state across separate remote environments.

The useful test for managed agents is not the demo task. It’s the teardown packet: files touched, tools called, cost, skipped steps, and what needs human approval before the next run.

google hosting my agent's messy terminal history on their own metal is bold

The remote Linux environment is the real unlock here, since tool reliability and reproducible state matter as much as the model call itself.

@readwise save

W update

@grok codemode in an isolated V8 edge container. Am I wrong?

b-but we were making it too oh my word you know what, hell yeah. google's killing it this year :D

Markdown won the agent config war, and nobody noticed

@grok did they copy Claude managed agents?

Background execution is where personal AI starts feeling less like chat and more like help. The product challenge is making the work visible enough to trust without making the user manage every step.

Managed agents with a Linux sandbox out of the box is a clear move to own the agent-to-production pipeline, not just the inference layer.

One API call for a full agent is wild.

remote Linux env in one API call is a big deal. less plumbing, more actual agent testing

Google's AI is becoming increasingly sophisticated.

Google is moving fast

All the plans lower than ultra are not worth it

remote Linux by default is a big step. the trust story gets harder once those environments start holding real repo context

shit version

Thank you for changing the world for us

@garrytan Would be interesting to make GBRAIN version / GBRAIN agents with this !

"remote Linux environment" means a box you can't SSH into. Google charges you to debug things you can't see.