Computer use, the browser tool, the Skills API, and the Files API are now generally available on the Claude Platform. Automate work in applications that have no API with fewer round trips per task, and build Claude Managed Agents on versioned skills and reusable files.

Claude now takes several actions per turn (click, type, key, screenshot) instead of one per round trip with `computer_toolset_20260801`. Early-access customers saw 20-40% fewer round trips per task, so tasks finish faster and cost less.

Browser automations that target pixels break when the layout shifts. `browser_toolset_20260801` gives Claude the page structure, so it acts on element references plus coordinates. Docs:

The Skills API and Files API are the building blocks of Claude Managed Agents. Skills API lets you upload your team's procedure once, version it, and pin each request to a `version_id` or `latest`. Docs:

The Files API lets you upload once and reuse the `file_id` across requests. Today’s update adds - `expires_in_seconds` - 5x higher rate limits (500 RPM) - 1 TB per org. Blog:

When Anthropic gave us early access to the new computer use tool, we immediately tested it on our healthcare computer use flows at @asteroid_inc . The numbers speak for themselves: - 32-52% fewer model calls - 25-32% lower cost per task 100% completion on every workflow, up from 77% We wrote a technical deep dive with the tips and tricks how to squeeze the most out of the Claude computer use models!

ChatGPT mogs ur computer use. If there’s one thing I can bet on not working, it’s Claude’s computer use.

Your computer use is the worst out of everyones, Claude can actually build better computer use for my agent app than it can for itself

Now can we get a weekly reset to try it out?

Automating apps without APIs is a huge leap, but the real bottleneck's gonna be latency on those browser tool round trips. If the UI shifts by a few pixels, does the agent hallucinate the button location?

Would be nice If you guys had a look at the memory. Normal Claude chat has no idea what tasks are being done inside of Cowork and Code. Would be nice to be able to use them all connected with a shared memory style so they know what stage of the project you on and you don’t remind

Most agent demos die on cost, not capability, since every single click burns a screenshot. Versioned skills is the sleeper feature here, it's the difference between an agent and a prompt you keep rewriting.

tf is the files api? didn't even know this was a thing

The "apps with no API" part is the whole thing honestly. Half my week is still reading a number off one ancient screen and typing it into another one two feet away.

does computer use cache “actions” it takes on applications? is this done per user, session, or at a broader claude level? how does it work under the hood is it just spitting out bespoke playwright / selenium scripts? @grok

We built a computer use agent that can actually navigate and use any software significantly fast by using native OS accessibility options @vovy_ai (

Maybe next week when my usages reset…. But likely won’t get that far before I’m capped again.

making agents more capable across real world workflows is a big step forward

The part that matters here is computer use going general. It means the model can click through a tool that never offered an integration, so what you can automate stops being limited to whatever happens to have an official plugin.

limit reset to party? 🥳🥳🥳🔥🔥🔥

Every 2-3 hours, there is an update from Anthropic. This is sick!

@bcherny Can we now finally automate ourselves out?

Computer use going GA is big news 🚀

Multiple actions per turn is the flashy bit. Versioned skills are the part future on-call engineers may actually thank you for.

pps with no public api rn: 👀

Bro is back 😭😭🔥

versioned skills let us manage agent tools like immutable CRDs

More automation, fewer friction points. The real question is whether one model's answer is enough before you act on it.

Computer use gets the headline, but Skills is the part I'd watch. We build with agents whose skills are just files on disk, so we can read, edit, and review them like any other code. You can see what your agent knows instead of hoping it's in the prompt: what's in a chat window disappears when the session ends, and what's in a file is still there tomorrow.

super excited to try this

Night Shift reset @ClaudeDevs

Thank you for this Claude. I am bit disappointed with claude recently..lets see whether it can help me again or not. :)

Improve your product. One message and my limit is reached. Is this what I'm paying 20 dollars for?

Can you make Managed Agents use subscriptions tokens do we don’t need to switch to GrokBot? And can you make Computer use available on Windows so we can use it for QA and can move away from Codex Computer Use?

fuck you you evil company, ceo in bed with epstein wife yuck

@lemhamza

Computer use going GA changes what's actually automatable in production.