Group 1

The agentic pipeline evolves in the dark

The team cannot freely test alternatives, compare them under identical conditions, or calculate the full cost of an accepted result.

Group 1 conclusion

The pipeline evolves without feedback

There is no room for free experimentation, no shared benchmark set, and no full cost for a result. Processes therefore evolve through intuition and personal habits rather than as a measurable engineering system.

Running example · AI review

The comparison should not be “bot or no bot,” but versions of the reviewer pipeline run on the same set of pull requests: measuring useful findings, noise, tokens, and human review time.