Why Daniel keeps working when you close the tab

Most AI coding tools run in the page you started them from. Close the tab, let the laptop sleep, or lose Wi-Fi on a train, and a 20-minute task stops halfway. For real engineering work that isn’t good enough, so we moved Daniel’s missions to the server.

How it works

When you start a mission, it is saved on our servers and a worker starts advancing it. Each worker runs for up to two minutes, saves its progress, and hands off to the next one. The page you’re looking at only watches. Close it and reopen it tomorrow, and it catches up from where the mission is.

If the chain of workers is ever broken, the next time anyone looks at the mission (or the daily check runs) it notices the mission has gone quiet and restarts it from its last saved step.

It recovers by itself

Long tasks hit ordinary problems. Daniel now treats three of them as things to wait out rather than failures:

• GitHub sign-in expired. The mission pauses and tells you. As soon as you sign in again, it resumes on its own.
• The AI model is busy. It retries automatically after a few minutes.
• The model account is out of credits. It retries every 15 minutes until the account is topped up.

In each case Slack or Teams tells you what it’s waiting for, so nothing sits silently.

You stay in control

Pause, stop, approve a plan or answer a question at any time, from any device. If a worker is in the middle of a step, your instruction is queued and applied the moment that step ends.

The result is simple: you give Daniel an issue and walk away, and you come back to a tested pull request.