Skip to content

Knowledge base

Quality gates

Why quality gates matter more once agents write the code, and how a gate differs from an instruction the agent may ignore.

Written by

Updated:

Glossary 33

A quality gate is a check with a fixed rule and a binary result: the build compiles or it does not, the tests pass or fail, coverage is above the threshold or below it. A coding agent writes code faster than people can review it, so these checks decide what all that output turns into.

Why do quality gates matter more with agents?

An agent optimizes to pass whatever checks it meets, so weak gates turn high volume into unusable output. Gates and validation have to grow faster than generation. The details are in Software Factories and Harness Engineering and The surrounding quality system.

How is a gate different from an instruction?

An instruction to run the linter sits in the prompt, and the agent can forget it. A gate is tied to an event and runs regardless. Git hooks gate the commit for every author, human or agent. Agent hooks gate the agent’s own actions mid-edit, and CI runs the full suite on push as the last gate. Each catches a different class of fault, and the earlier a gate fires, the cheaper the fix. Protect the gates themselves as well: an agent under pressure to pass will edit the lint configuration or reach for --no-verify unless the repository forbids it.