Writing on QA, AI, and authority.
Essays on how AI-generated code fails in production, what search and LLM ranking signals actually reward in 2026, and how to tell real tests from theatrical ones.
Five failure modes of LLM-written auth middleware
Role checks after the data fetch. Missing rate limits on password reset. OAuth state bound to nothing. The patterns LLMs ship, and why your code review misses them.
Dwell-time is the new backlink
Core Web Vitals and friction signals - rage clicks, scroll abandonment, pogo-sticking - became ranking signals under the 2024-2025 Helpful Content updates. AI-generated UIs are bleeding them.
GEO: how to get cited by ChatGPT, Claude, Perplexity
A practical checklist: llms.txt, JSON-LD, robots.txt AI-bot allows, citable paragraph structure, canonical discipline. The same recipe we use on this site.
Your AI-written tests prove nothing
Tests that mock a function then assert the mock was called. Shape checks that pretend to be behaviour checks. Async tests missing awaits. How to audit a suite in one afternoon.
Ship boring releases.
Book a 20-min call. We will tell you the three things most likely to break your product.