Field notes from teams shipping AI into production
Practical writing on agentic AI, application modernization, and product engineering. No hype, no recycled trend pieces. Just what worked, what broke, and what we would do differently next time.
AI-generated code lands in the technical debt quadrant nobody chose. Where it forms, why review can’t catch up, and the generation-time rules that prevent it.
Reviewing AI-generated code is not one read done harder. It is five: functional, regression, security, budget, maintainability. The framework, and how to run all five without slowing merges.
The test suite passed and CI was green, yet production broke. Why AI-assisted code slips past pipelines built for human-paced change, and what verification has to become.
AI speeds up coding, not delivery. What actually gets faster, where a software estimate really goes, and how to set a timeline you can take to your board.
Most AI workflows treat the code as if it were the theory. On a brownfield system it is not. Codebase archaeology is the discipline that produces the written understanding an agent needs before it edits.
An AI coding agent will write confident code against assumptions it invented. On a 12-year-old codebase, that is where the bugs come from. What the agent has to understand first, and how to brief it.