AICoder

Appearance

Coding sessions ​

A session is one run of an agent on a task. The session panel is where you watch it work and step in when needed.

The live session Monitor view

The three tabs ​

  • Monitor — the current step, the agent's todo list, and its live thinking. The at-a-glance view of what the agent is doing right now.
  • Log — the full streaming output: every tool call and decision, rendered as command blocks.
  • Diff — the live diff of changes as the agent writes code.

The Log tab showing command blocks

Answering questions ​

If the agent gets blocked, it asks a question. A question box appears in the session (and the session moves to Needs you in the Queue). Type your answer to unblock it — the agent resumes with your reply in context.

Follow-ups ​

You can send follow-up instructions mid-session — for example "also update the tests" — without restarting. Suggested next steps appear as a follow-up bar.

Parallel agents ​

When a task has several agents running, a rail on the left lists them so you can switch between concurrent agents quickly.

Controls ​

  • Close session — leave the session running in the background (it stays in the Queue / Workbench).
  • Stop / kill — end a session that's stuck or going the wrong way.
  • Resume — pick a paused session back up with its remembered context.

What happens when it finishes ​

A clean exit triggers the automatic review pipeline. The finished run then appears under Needs you in the Queue, ready for you to review the diff and ship.