Across 400,000 Claude Code sessions analysed, non-developers complete their tasks successfully 89% of the time, against 91% for software engineers. The gap that matters separates people who know their field from people who do not, far more than it separates technical from non-technical profiles.
The facts
- Sessions run by users who are experts in their field generate 2.4 times more actions than novice sessions.
- Observed split: users take 70% of the planning decisions, the AI takes 80% of the execution decisions.
- 19% of struggling novice sessions are abandoned, against 5 to 7% for users experienced in their field.
What do these 400,000 sessions tell us about AI adoption?
Two points separate non-developers from engineers: the technical barrier that defined thirty years of computing is coming down. What replaces it is more demanding than it looks: knowing precisely what you want to obtain, telling a correct result from a plausible one, and breaking a problem down into steps. All of this comes from domain expertise rather than from command of a programming language.
The observed split of decisions, with humans deciding 70% of the planning and the AI carrying out 80% of the actions, describes a division of roles that owes nothing to chance: it is the one that works. The AI produces, the human steers. And the near-quadrupled abandonment rate of novices in their own field confirms the corollary: putting a powerful tool in the hands of someone who does not know their subject produces no expertise, only faster dead ends.
For management teams preparing their training plans, the message is counter-intuitive and valuable: a course on the tools is rarely the best AI investment on its own. Formalising the company's domain knowledge, what its experts know and nobody has yet written down, often pays more, because that is exactly the material AI multiplies.
AIxH's view
Training a team on an AI tool without working on its domain knowledge is not enough. That is the principle behind our training programmes and our audits: the tool comes second. The best results we see at our clients always come from teams whose know-how was valued first.