Jason Turner recently gave a talk at a C++ conference about best practices for AI tool use. Given that most of us are now either mandated or strongly encouraged to use these tools, his timing couldn’t be better. But after watching the full presentation and comparing his recommendations against the latest research, I found myself nodding along to some parts and shaking my head at others.

Turner’s safety-first mindset is absolutely correct. His observations about AI tools disabling warnings and gaslighting developers about test failures ring true for anyone who’s spent serious time with these agents. But his recommendations also reflect an overly conservative approach that might have made sense in early 2024 but hasn’t kept pace with how rapidly these tools are evolving.

Let me walk you through what he got right, where he missed the mark, and what he completely forgot to mention.