What happens when AI handles the hard parts of a problem? You stop trying as hard.
A study on AI assistance and task persistence found that access to AI tools measurably reduces how long people push through difficult work before asking for help. Persistence here means the tendency to keep working on a challenging problem rather than seeking external help - it's a proxy for how much productive struggle actually takes place.
The finding is intuitive once stated. When a solution is one prompt away, the internal cost-benefit calculation changes. The cognitive friction that normally builds skill gets bypassed at the exact moment it would otherwise kick in.
What gets lost in the handoff
For daily AI users, this effect is specific to where AI is available. A developer who hands every debugging problem to an AI model never develops the habit of reading error messages and forming hypotheses - the slow, frustrating work that builds diagnostic skill. A writer who breaks every creative block with ChatGPT never learns to push through resistance, which is a separate skill from writing itself.
The study doesn't suggest AI makes people globally less capable. Persistence declines specifically on tasks where AI is available. What that means in practice depends on which tasks you're routing to the model.
Using AI without quietly losing ground
Practitioners who avoid this pattern tend to follow a consistent approach: AI handles tasks already within their competence - formatting, boilerplate, information lookup - while they preserve their own effort for work at their growth edge. The model compresses time on the known; they struggle with the unknown.
A second approach: attempt the problem first, then use AI to check or improve the output. The struggle still happens. AI shortens the iteration cycle after it.
The risk here isn't sudden incompetence. It's a slow, invisible narrowing of the things you're willing to struggle with. Skills atrophy quietly when you stop exercising them, and AI makes it easy to stop without noticing - because the work still gets done, just not by you. Pointed at tasks below your skill level, AI multiplies your output. Pointed at your growth edge, it quietly removes the ceiling-raising that would otherwise have happened.