I teamed up two AI tools to solve a major bug - but they couldn't do it without me (www.zdnet.com)

A developer wrestling with a mysterious slowdown in his WordPress plugins used two AI tools to find and fix the issue: OpenAI’s Codex and ChatGPT Deep Research. Codex could write code and build a verbose diagnostics logger but struggled with big-picture, cross-release diagnosis. Deep Research, which could analyze full repo and distribution zips, compared the working v3.2 release with the problematic v4.0 and found the root cause: the main plugin was checking the robots.txt status on every page request. Those synchronous checks blocked the PHP interpreter and, under real traffic, effectively froze the site. Armed with that diagnosis, the author had Codex implement a focused fix—check once at startup, cache the status, and provide a manual “recheck” button—resolving the outages in production. The story highlights practical lessons for AI/ML development workflows: different AI assistants have complementary strengths (Codex for targeted coding, Deep Research for repository-scale diagnostics), but human testing, version-aware reasoning, and orchestration remain crucial. Technically, it’s a reminder that synchronous IO in request paths can cripple PHP-based sites and that simple caching and administrative recheck controls are effective mitigations. More broadly, teams should pair AI tools with regression-aware analysis and realistic load testing; that combination—AI diagnosis + AI implementation + human oversight—can speed development but won’t replace careful testing and product management.