Week 9: Background agents

This week asks what you can accept from an agent that worked while nobody was watching. The lecture gives one bounded request to a foreground session and to an unattended one and checks both results against the same unchanged cases. It then separates the three boundaries of parallel work, and walks an issue-to-pull-request pipeline stage by stage to name the gate that blocks each failure and who controls it. Every step pairs something generated (a candidate, a branch, a pipeline design) with a check you run yourself. The duration example is course-written and worked by hand; none of it is a model run.

Start with the notes. The second session is a guest session or the studio, where you run one small task from your project or current contribution twice from the same base and decide what to accept. Then use the self-check and the readings. The slide source follows the same sequence. Contribution 3 is due this week.