AI is making your team faster. It may also be making them less capable.
That tension sits at the heart of every serious conversation about AI integration in the workplace. The productivity numbers are real. So is the risk. Understanding and designing for both is now a leadership responsibility, not a technical one.
Productivity Gains Are Real, But So Is the Hidden Cost
The evidence for AI’s productivity impact is compelling. Research shows AI-based conversational assistants increased the number of issues call centre workers resolved, consultants using AI complete more tasks than those who didn’t, and software developers with AI coding assistants complete tasks faster, with novice programmers benefiting most.
These are significant gains. But they tell you what gets done faster, not what gets learned.
When people use AI to complete tasks that would otherwise require them to think, reason, and solve problems, they often skip the cognitive effort that builds lasting capability. Medical professionals who rely on AI for diagnosis risk not developing the visual pattern recognition needed to identify conditions independently. Knowledge workers who regularly delegate thinking to AI report lower cognitive effort, reduced confidence, and associations with weaker critical thinking skills.
The risk is not that your team becomes slower. The risk is that they become shallower.
AI Assistance Does Not Automatically Accelerate Learning
A 2026 study by Shen and Tamkin provides some of the most direct evidence yet on AI’s impact on skill formation in professional settings [1]. The researchers ran randomised experiments with software developers learning a new library, some with AI assistance, some without.
Developers who used AI assistance showed measurably worse conceptual understanding, code reading, and debugging ability when tested independently afterwards. Those who fully delegated coding tasks to the AI completed more work during the experiment, but they failed to learn the library they were ostensibly practising with. Compounding this, the study found no significant acceleration in overall task completion time with AI assistance on average. Time saved on execution was largely offset by time spent querying the AI.
The most practically useful finding was this: not all AI interaction patterns produce the same outcome. Six distinct patterns were observed, and three of them preserved learning even with AI assistance. The differentiating factor was cognitive engagement. Developers who asked the AI to explain concepts, explore alternatives, or walk through reasoning retained their skills. Those who simply asked the AI to do the work did not.
How your team uses AI matters as much as whether they use it.
Junior Staff Face a Different Problem Than Senior Staff
For employees with deep domain expertise, AI is genuinely additive. Their knowledge provides the context to direct AI effectively, identify errors, and extract real value. They can delegate more because they have more to draw from.
For junior staff, the picture is more complicated. The tasks AI now handles – first drafts, basic research, routine analysis, initial code – are precisely the tasks through which people historically built their understanding. Remove those tasks, and you remove much of the learning pathway.
This is a systems problem, not an individual one. You get what you incentivise, and if the incentive is output without effort, you will eventually run short of the expertise that makes that output valuable. Organisations that adopt AI without redesigning how junior staff develop capability will find their senior pipeline hollowing out over time. Not immediately, but in three to five years when today’s novices are expected to operate independently.
Capability Must Be Designed Into AI Workflows, Not Assumed
The solution is not to limit AI use. It is to be deliberate about how AI is integrated into development and workflow.
Separate productivity tasks from learning tasks. Not every task needs to be a learning opportunity, but the ones that are should be protected. When a junior employee is developing a new skill, the workflow should require them to work through the problem first before AI assistance becomes available.
Design for cognitive engagement. Encourage staff to use AI for explanation, exploration, and review, not just task completion. Ask them to attempt problems independently, then use AI to check, challenge, or extend their thinking. This mirrors the interaction patterns Shen and Tamkin identified as preserving skill development and used in traditional senior mentor relationships.
Build evaluation skills explicitly. AI output can be wrong. Knowing when to trust it, when to question it, and when to override it requires genuine domain knowledge. That capability needs to be developed intentionally and cannot be assumed to emerge on its own.
Vary how AI is used. Teams that use AI only for task delegation will learn less than teams that use it for conceptual exploration, structured critique, and pattern checking. Breadth of interaction pattern matters.
Reassess workflows with capability in mind. Most current workflows were built for human execution. Integrating AI into them without revision often means AI handles the developmental steps and humans handle review, which is backwards for skill formation. The structure of work needs to be rethought, not just the tools within it.
The Leadership Question
AI will make your organisation more productive. That outcome is largely settled. The question that remains open is whether that productivity will be built on a foundation of growing capability or a gradual erosion of it.
The organisations that navigate this well will treat AI integration as a capability design challenge, not a tooling decision. They will ask not just “what can AI do for us?” but “what do we need our people to be able to do, and how do we ensure AI supports that rather than substitutes for it?”
Preparation determines performance. Not the tools available in the moment, but the depth of capability built over time through deliberate practice. That principle does not change because AI arrived. If anything, it becomes more important because AI makes it easier than ever to skip the work that builds the capability you will eventually need.
References
[1] Shen, J. H., & Tamkin, A. (2026). How AI Impacts Skill Formation. arXiv preprint arXiv:2601.20245. https://arxiv.org/pdf/2601.20245