When AI Clicked: Why Interrogating Code Matters More Than Vibe Coding

Summary

Ewan Valentine describes how his initial use of AI as a productivity tool left him feeling detached from the thinking process. As AI began generating pull requests, features, and bug fixes, he became increasingly concerned that he did not fully understand either the code or the larger codebases into which it was merged. He argues that extremely large productivity gains could increase codebase entropy if understanding fails to keep pace with generated code. His response was to interrogate every obscure output by asking why a method was chosen, how it addressed the original issue, what would happen in edge cases, and whether visual examples could clarify it. This reduced his reported productivity gain to about 20%, but shifted his time toward reading, debating, and challenging assumptions. He says the approach has improved his comprehension, exposed frequent and sometimes poor AI decisions, and restored the intellectual effort he associated with engineering. After briefly fearing that AI had replaced him, Valentine concludes that engineers who prioritize deeper understanding over unchecked vibe coding may become better engineers, while acknowledging that this view is provisional.