AI Coding agents

Hand over a task. Get back a pull request.

Say “implement T-42”, or press Run coding agent on the task. The agent works on its own branch in the background, asks before it does anything unusual, and opens a pull request for your review.

Leo

Docs are updated for the new checklist. Two screens went away entirely.

Tomas

Nice. I moved the two tasks that covered them to Done.

Priya

Pricing page is live. Copy went through two rounds with Leo.

Screenshots are in the launch doc if anyone needs them.

Today

Tomas

Week one numbers are in the launch doc. Activation is up 12%.

Priya

Support inbox is at zero. First time since launch.

Nadia

Onboarding rework is live. The checklist is down to four steps.

🎉 2 ❤️ 1

@Leo step three next?

Leo

Huge. Step five was where everyone dropped off.

Press '/' for commands

Every piece of work, in one list.

Illustrative demo · build time condensed

How it works

  1. Hand it over

    From chat, from ⌘O, or with Run coding agent on any task. Pick the model.

  2. It works on a branch

    Reading, editing and running commands are approved from the start. For anything else, it asks.

  3. Review the pull request

    The run ends in a pull request on its own branch, linked to the task.

Work you can review.Not work you have to watch.

  • Asks before it strays

    Allow once, allow for the run, or decline. A question left for ten minutes stops the run and saves its work.

  • A branch each

    Hand over several tasks and each run works on a branch of its own.

  • Local, on your plan

    On the desktop app, run it with your own Claude Code, Codex, Grok or Hermes instead of credits.

  • Answers the review

    Press Address on a review and the agent works on the PR’s branch and replies in each thread.

  • No repo, still useful

    Without a repo, a run works with the web and a scratch shell and hands back a written result.

  • Spend under control

    Admins can switch off credit-paid runs and cap how many turns a run can take.

It asks before it strays.

Reading, editing and running commands are approved from the start. When a run wants something outside them, like a web search, it stops and asks in the run, and whoever answers decides for the rest of it.

Good to know

  • Nothing merges on its own. Every run ends in a pull request a person reviews.
  • Cloud runs use credits. Local runs on the desktop app use your own subscription.
All AI features