What's this about?
Lars Faye's argument, following his earlier "Agentic Coding is a Trap", is what he calls the expert novice problem: the developers benefiting most from these tools are those with decades of pre-AI experience, while newcomers are told to accelerate with tools that require exactly the expertise they have not had a chance to build.
What exactly is "skill atrophy"?
The claim is that expertise comes from doing the work, including the frustrating parts. If the tool handles the friction, the learning that friction produced never happens. Faye's specific twist is the paradox that the skills needed to supervise an AI agent well are the same skills that using the agent prevents you from developing.
What's the community saying?
HN 542 points, 535 comments
- A widely-endorsed report from inside enterprises: leadership mandates that writing code by hand means doing it wrong, output volume rises, and the consequences are arriving later.
- The standard rebuttal, delivered flatly — "and math skills are going to collapse from calculator reliance."
- A defence of friction as the engine of learning, with the muscle-atrophy analogy.
- A practical middle path with real support: guided coding — an editor with an integrated model where you still write the code — as distinct from headless agentic coding, and underdiscussed.
- A tangent on compensation and whether a correction is coming for developers who mostly supervise agents.