An InfoQ article proposes agentic fitness functions as a way to extend evolutionary architecture beyond rules that are easy to measure with deterministic checks.
Traditional fitness functions work well for constraints such as dependency direction, contract shape, latency budgets, security posture, and policy checks. The article argues that AI agents can help with evidence-bound but judgment-heavy concerns, including boundary fidelity, semantic contract drift, and stale architecture decision records. In that setup, versioned rubrics and calibrated feedback loops turn architectural intent into continuous review rather than occasional governance meetings.
The useful distinction is that agents are not presented as replacements for hard checks. Deterministic rules should remain the primary enforcement mechanism where the desired condition can be measured directly. Agentic checks are better suited to areas where teams already need expert interpretation. That makes the idea promising, but also dependent on careful rubrics, audit trails, and human review when architectural consequences are significant.