newsletter.techworld-with-milan.com
|
ksl
|
|
Milan Milanović, a startup CTO with 20 years of experience, argues that AI coding tools are making the wrong skill faster. He cites a METR study where developers believed AI made them 24% faster but actually worked 19% slower, and GitClear data showing AI-assisted code produced 4x more duplication and higher churn. His checkout case study drives the point home – engineers assumed slow queries caused user abandonment, but the real problem was form complexity, fixed in 40 lines with an 18% completion lift. The argument isn’t new, but it lands differently now that Claude Code and Copilot can generate thousands of lines in minutes. Anthropic, Amazon, and Google have all published internal research reaching similar conclusions – problem framing and architectural judgment are where senior engineers earn their compensation, and those skills remain stubbornly hard to automate.
