Dev Tutorial Introduces 'Progress Contract' to Stop AI Coding Agents from Spinning in Place

A software development tutorial published on DEV Community introduces a bootcamp-style lab designed to expose a common flaw in AI coding agents: appearing busy without making real progress. The lab requires agents to track a 'remainder' — a measurable, shrinking count of unfinished work such as failing tests or unresolved files — at every iteration. If the remainder does not strictly decrease each loop, the run is graded as a failure, regardless of how much output the agent produces. The exercise uses Node.js, a frozen test oracle, and an append-only NDJSON log to enforce accountability and prevent agents from gaming their own exit conditions. The author argues that without such a progress contract, agent transcripts resemble a diary rather than proof of forward movement.