AI agents can help investigate failed builds, fix flaky tests and automate other development tasks. But setting up those agents has required more Pipelines configuration than it should. Agent-powered steps often need different compute, permissions and runtime settings from ordinary build and test steps.