I see many teams rushing to plug AI Test Agents directly into CI/CD and expecting magic.
The core truth: most agents are instruction-driven. They generate or update specs when guided — not continuously.
So the real question becomes:
- If that's the case, why put them in CI at all?
- Why not run them locally → review → commit → let CI run deterministic tests?
In many teams, local usage is enough.
CI integration starts making sense when you need scale, consistency, governance, or controlled workflows across multiple repos and teams.