Assign a task to the AI, press “Run coding agent”, or just say “implement T-42” in chat. A cloud agent clones your repository into a sandbox, works the ticket on its own branch and opens a pull request. You review it the way you review everything else.
A small team’s backlog is full of work nobody disputes and nobody has an afternoon for. Meanwhile the AI that could help is in a different tab, with no idea what T-42 is, no access to the repository, and no way to hand you back anything but a snippet.
Describe the problem, paste the file, paste the answer back, discover it assumed a framework you don’t use.
Copy fixes, empty states, a flaky test. Each one is twenty minutes that never arrives.
Code that arrives as a message hasn’t been diffed, hasn’t run CI and can’t be commented on line by line.
Assign it to the AI, use the task’s “Run coding agent” button, or ask in chat, by ⌘O voice, or in a huddle. Pick which model runs it.
Your repository, cloned into an isolated cloud container, on its own branch — with the task, its comments and the workspace’s context as the brief.
It opens a real PR on GitHub. Your CI runs, your reviewers comment, and nothing reaches your default branch that you didn’t merge.
The task shows a live badge while the agent works and moves on when the code lands, so the column reflects reality rather than memory.
Chat, decisions, docs and recent commits are all in context, so “like we did for the invite emails” is an instruction it can follow.
Coding agents are on every plan, including Free — they spend AI credits, and every workspace starts with 1,000.
Settings → GitHub. One workspace, one repository — that constraint is what makes the context worth anything.
Free for up to 10 people, with 1,000 AI credits included. No credit card, no sales call, no onboarding deck.