Uncle Bob Stopped Reading AI-Generated Code

Get my Claude Code SKILLS for Clean Architecture: https://milanjovanovic.tech/templates... Want to master Clean Architecture? Go here: https://dub.sh/clean-architecture Want to master Modular Monoliths? Go here: https://dub.sh/modular-monolith Join the .NET Architects Club: https://www.skool.com/mj-tech-communi... Get the 2026 .NET Developer roadmap here → https://the-dotnet-weekly.ck.page/202... Uncle Bob says he no longer reads the code written by his AI agents. Reckless - or the direction software engineering is already heading? I unpack the viral claim and the part most reactions miss: this is not about blindly trusting an LLM. It is about deciding how much review the risk demands, then replacing line-by-line inspection with stronger constraints and verification. We cover: How application risk changes how much AI-generated code you should read How unit, integration, architecture, and Gherkin tests constrain AI coding agents Where manual QA, mutation testing, agent skills, and custom harnesses fit Why code coverage can reveal gaps without proving quality or correctness Why "all AI-generated code is slop" is too simplistic What happens to the developer’s role when agents produce most of the implementation I also explain how I apply this thinking while building Katabench and why code is only one artifact in the larger job of engineering reliable systems. This is for software engineers using Claude Code, Codex, or other AI coding agents who want more productivity without blindly shipping whatever the model produces. Check out my courses: https://www.milanjovanovic.tech/courses Read my Blog here: https://www.milanjovanovic.tech/blog Join my weekly .NET newsletter: https://www.milanjovanovic.tech Chapters