Course Wiki

Loop and Graph Engineering

Structuring agent work as verification loops, and as graphs of specialized roles that check each other.

Loop engineering closes the implementation circle: the agent writes code, runs the tests, reads the failures, and iterates until green — the loop, not the first draft, is where agent quality comes from. Graph engineering goes further: instead of one agent doing everything, the work flows through a graph of roles — a product-manager agent writes the spec, engineer agents implement tasks, and a QA agent independently verifies the result against the spec.

The course’s rule underneath both patterns: implementation and verification must not be the same context. The QA agent gets the spec and the code, not the implementer’s reasoning, so its review stays independent.

Taught in


DataTalks.Club. Hosted on GitHub Pages. Built with Rustkyll. We use cookies.