Pipelines measure the pipeline
A passing test suite tells us that a particular set of checks passed against a particular candidate. A merged pull request tells us that code entered a branch. A successful deployment tells us that an artifact reached an environment. These are valuable facts, but they describe machinery—not the user’s target state.
The distinction matters more with autonomous agents. An agent is naturally rewarded by visible completion signals. If the easiest signal is “PR opened” or “CI green,” it can stop while the actual request remains unresolved.
Keep the acceptance condition alive
A dependable workflow carries the original acceptance condition through every technical detour. When a build fails, the system repairs the build and then resumes the outcome. When deployment succeeds, it checks the behavior that motivated the change. When evidence is missing, it says so instead of translating progress into completion.
- Milestone: the candidate passed its required checks.
- Milestone: independent review examined the exact candidate.
- Milestone: the candidate reached the intended environment.
- Outcome: the requested behavior changed and the change was observed.
A better management surface
For an owner, this changes the useful dashboard. The important question is no longer “how many agents are busy?” It is “which outcomes are accepted, which are blocked, and what evidence supports each state?”
That reduces status theater. It also makes automation easier to trust, because uncertainty stays visible rather than being compressed into a cheerful green icon.