Heavy AI use risks shrinking your team's problem-solving skills

Relying on AI for routine answers can save time now but raises the risk of long-term loss of staff capability and increased training cost.

What actually changed

Eric So’s new book, published this month by W.W. Norton, compiles recent research and industry analysis to argue that extensive reliance on AI can reduce people’s ability to learn and retain complex skills. The book notes AI can speed work and serve as domain-specific tools in useful cases, but So and the studies he draws on conclude that learning improves when people struggle with material, fail, iterate and thereby build durable thinking skills. The reporting cites an Anthropic analysis showing that roughly half of college-student queries were what the company classified as "low-effort queries" for simple answers—evidence that many users default to quick prompts rather than sustained problem-solving.

Who it affects

Small teams, freelancers and managers who delegate knowledge work are most exposed. Where routine answers replace hands-on problem-solving, teams may appear more productive in the short term while losing the deeper skills that reduce future risk and external dependency. The Anthropic finding specifically involves college students, but the pattern—defaulting to quick-answer prompts—translates to any group that uses generative models as an easy substitute for iteration and practice.

What it costs or what it replaces

Pricing: the source does not state pricing for any tool.

What it replaces: according to So’s synthesis and the cited research, heavy AI use often replaces the iterative practice that strengthens learning—working through initial failures, refining solutions and internalising techniques. The immediate cost is weaker long-term retention and lower problem-solving ability; the likely indirect costs for a small team include more time spent on rework, higher onboarding and training needs, and greater reliance on external tools or consultants.

What it saves: the source acknowledges AI can complete tasks faster and serve as effective domain-specific tools in some cases. That short-term time saving is documented as a benefit, but So’s argument frames it as a trade-off against degraded learning if used as a default.

What we don't know

• Whether the Anthropic analysis covered professional users or only students.
• How large teams experience the same effect compared with individual learners.
• The rate at which lost skills can be recovered through targeted training after prolonged AI reliance.
• Any cost figures or pricing for the models or tools mentioned (the source does not provide pricing).

What to do next

1. Audit routine prompts your team runs today and tag those that seek simple, single-step answers; stop using generative AI for tasks that can be turned into short checklists or reference lookups and reserve it for genuine productivity gaps.
2. Create regular exercises where staff must solve problems without AI assistance—short, timed tasks that force iteration and failure—so learning is preserved while still using AI selectively for domain-specific work.
3. If you are considering switching primary workflows to generative models, pilot for a fixed period and measure rework, onboarding time and problem-resolution rates before rolling out across the team.