Optimizing a slow API with Redis caching is squarely in AI's wheelhouse.

This is a well-understood class of performance problem — N+1 queries, missing indexes, and cache-aside patterns are textbook. The solution structure is highly repeatable across similar stacks, which strongly favors automation.

Success criteria are concrete and measurable: response time under 500ms, Redis TTL-based invalidation, and a cache-warming strategy for high-traffic SKUs. There's little room for subjective interpretation of 'done'.

The agent needs the actual Express route code, PostgreSQL schema, and Redis connection config to produce accurate output rather than generic scaffolding. Without these, the code will be structurally correct but may not integrate cleanly.

Badly implemented caching can cause stale data bugs or cache stampedes, but these are detectable in testing before deployment. The risk is real but reversible — no data loss or security exposure from a flawed first draft.

Decisions like TTL duration, cache key design, and index selection are driven by well-established engineering heuristics, not taste or relationship context. A senior engineer should review the output, but the judgment bar is low.