How do you let your AI work on its own without it running forever, spending too much, or making changes you cannot undo?
- Module
- 03/05
- To read
- ~9 min
A team left an agent running all weekend in a script that restarted it until the test suite passed. One failing test depended on an outside service that was down. What do you think they found on Monday?
In the module you make your guess first, then read what happened and why.
Budget before you start
Save points after every verified step
Progress is measured by your code, not the agent
Every run ends in a named state
Rules for AI runs you are not watching
A spec for your own app, as a text file you download. The module also gives you the prompt for Claude Code, Codex or Antigravity, a checklist to check what it built, and recommended skills.
- Job queues and dead-letter queues
- Cost control