Five GitLab Duo AI agents handle post-coding work in a real Python service pipeline

A developer tested five GitLab Duo agents covering planning, review, fixing, security and release on a small Python service called ro-cited-answers. The agents planned a tracing feature in about two minutes and drove a pipeline race diagnosis, while cutting container scan findings from 166 to one. The agents were restricted to opening merge requests and writing notes, and they could not merge or deploy, so the developer did the merging and made one mistake doing it. The project's pipeline covers verification, packaging, security scanning, staged release with rollback, configuration drift checks, and production monitoring, with deployments authenticated through Workload Identity Federation instead of service account keys. The author presents the account, including two mistakes, as a candid look at how much post-coding work AI agents can take on and where human oversight remains necessary.