Simon Willison: AI coding agents won't end software engineers

Willison's post is a response to a familiar moment in software engineering: a coding agent finishes in an hour work that would have taken a week, and the engineer who used to do that work feels the ground shift. His argument is that this feeling is not the end of the profession but a stage. Willison places the change in the longer history of software engineering, pointing out that the tools and languages engineers use have rarely stayed stable beyond about a five year time horizon, and that anyone who chose software development as a passion opted into frequent radical change from the start.

The distinction he draws is between the narrow skill of turning an exact specification into decent code and the larger set of problems a software engineer faces. The first is no longer unique, he says, but the second remains, and experience with the wider set is what lets engineers master the new tools. That framing suggests the value shifts from producing code by hand to directing and judging the work of agents, a change that may be harder for people who want their profession to stay exactly as it is.

The stakes hinge on whether engineers treat agents as a replacement for their identity or as a tool their existing depth lets them use well. Willison's own claim is that those who come to terms with the shift can provide value and execute at a level far greater than newcomers who build software with agents but without that depth. It is a hopeful reading of the transition, though it rests on the engineer's willingness to keep changing.