AI can do most of this Django query audit, but someone still needs to run the benchmarks.

N+1 query detection and ORM optimization are highly pattern-driven tasks with well-established Django idioms. The same analytical approach applies across any Django codebase, making this structurally repeatable.

Success criteria are partially crisp — reduce query count, improve load time — but the acceptable threshold for 'done' (e.g., how many queries is acceptable, which indexes are worth adding) involves tradeoffs that depend on production context and team standards.

The agent needs full codebase access, Django Debug Toolbar logs, and ideally a runnable environment to validate metrics. If given static code and toolbar output, it can analyze and rewrite, but cannot independently produce verified before/after benchmarks.

Incorrect query rewrites could introduce subtle bugs — wrong prefetch paths, missing annotations, or broken related-object access — that may not surface immediately. Database index additions are low-risk but schema migrations require care. Errors are reversible but potentially disruptive.

Identifying which queries to optimize and how to balance query count vs. memory (e.g., large prefetch sets) requires some architectural judgment. A senior engineer should review the proposed changes before merging, especially for index decisions.