There is a quiet tension building inside organizations that have embraced AI tools broadly and enthusiastically. The tools work. They accelerate output, reduce friction, and handle volume that would otherwise slow teams down. The concern is not whether AI is useful. The concern is what happens to the people who never had to work without it.
This is not a nostalgic argument for doing things the hard way. It is a structural question about how expertise actually forms, and whether the conditions that produce deep professional judgment are being quietly eroded in the name of efficiency.
How Expertise Forms: The Role of Struggle, Error, and Repetition
Cognitive science has long distinguished between surface competence and genuine expertise. Surface competence means being able to produce an acceptable output. Genuine expertise means understanding why that output is correct, recognizing when it is not, and knowing what to do when the standard approach fails.
Expertise is built through a specific kind of experience: encountering problems without a ready answer, making decisions under uncertainty, receiving feedback, and adjusting. This cycle, repeated across many contexts over time, is what builds the mental models that distinguish a senior professional from a junior one.
The critical detail is that this process requires friction. A beginner who struggles to write a first analysis, gets it wrong, understands why it was wrong, and tries again is building something durable. A beginner who prompts an AI system, receives a polished output, and submits it has completed a task. These are not the same activity, even when the outputs look identical.
This is what most coverage of AI in the workplace misses. The conversation focuses almost entirely on output quality and speed. It rarely asks what is happening to the person producing the output, and whether their capacity to produce it independently is growing or staying flat.
The Scaffolding Problem: When Assistance Becomes a Ceiling
In education and developmental psychology, scaffolding refers to temporary support structures that help learners reach levels they could not reach alone. The key word is temporary. Good scaffolding is designed to be removed. As the learner grows, the support is withdrawn, and the learner internalizes what the scaffold once provided.
AI tools, as currently deployed in most professional environments, do not work this way. They are permanent, always available, and increasingly capable. There is no mechanism that withdraws assistance as a person develops. The scaffold stays in place indefinitely.
This creates a specific risk for early-career professionals. If a junior analyst always has access to an AI that can structure arguments, identify patterns, and draft conclusions, the analyst may never develop the independent capacity to do those things. They become skilled at directing the tool, which is genuinely valuable, but they may not develop the deeper judgment that comes from doing the underlying cognitive work themselves.
The problem compounds over time. A professional with five years of experience who spent those years working with heavy AI assistance may have a very different capability profile than one who spent those years working through problems manually. The gap may not be visible in daily output. It becomes visible in novel situations, in crises, in the moments when the tool cannot help because the problem has no precedent.
Why This Matters Beyond Individual Careers
The implications extend beyond any single professional’s development. Organizations depend on a pipeline of expertise. Senior professionals mentor junior ones, pass on judgment that cannot be written down, and eventually retire or move on. If the junior cohort entering the workforce over the next decade develops shallower expertise because AI handled the formative cognitive work, the pipeline narrows.
This is not a reason to restrict AI tools from professional environments. It is a reason to think carefully about how and when they are introduced, and for what tasks. There is a meaningful difference between using AI to handle high-volume, low-complexity work so that junior professionals can focus on harder problems, and using AI to handle the hard problems so that junior professionals never have to engage with them.
Organizations that think carefully about this distinction are likely to produce stronger professionals over time. Those that deploy AI uniformly across all tasks and all experience levels may find, a decade from now, that they have efficient teams with a notable shortage of people capable of operating without the tools.
In Short
AI tools accelerate output and reduce friction. That is genuinely valuable. The risk is that for early-career professionals, the cognitive struggle that builds deep expertise is also being reduced, and that reduction may not show up as a problem for years. Expertise forms through encountering difficulty, making errors, and building judgment through repetition. When AI handles that difficulty on behalf of beginners, the output improves but the person may not. Organizations that want experienced, independent professionals in ten years need to think now about which tasks their beginners should be doing without assistance, and why.