You can’t afford to lead agentic engineering from the sidelines

Engineering leaders who design AI operating models without hands-on experience risk building plans that fail on contact with real codebases. Drawing on firsthand experience leading AI adoption as Director of Engineering, the author argues that leaders must personally use agentic tools to understand where they genuinely help, where they shift cognitive burden back to engineers, and where teams aren't ready. Key points include: agents don't eliminate the need for human judgment, they just make bad judgment cheaper to execute; AI exposes pre-existing bottlenecks like slow CI/CD and vague product direction; engineers need product context to direct agents effectively; and organizations should prioritize shared learning over premature standardization of workflows.