AI is changing how expertise develops: faster than many organizations are redesigning for it.
Many of the activities that historically contributed to expertise development are among the first being automated. The disruption is not technological. It is developmental.
AI is making everyone faster. That is not the problem.
The problem worth paying attention to is this: the work being automated first often provided the earliest experiences through which expertise began to develop. Repetitive research, first-pass analysis, and routine drafting felt low-value. But they exposed professionals to patterns, ambiguity, and tradeoffs that contributed to later judgment. By judgment, I mean something specific: the ability to form a sound assessment under uncertainty, weigh competing considerations, and act on incomplete information. Not intuition, not instinct, but a capability built through repeated exposure to situations where the answer was not obvious. That development happened slowly, invisibly, and cumulatively. It did not feel like development at the time. It only became visible years later, in the quality of someone’s instincts.
The argument that follows applies most directly to roles where expertise has a significant judgment component: where the work requires interpreting ambiguous situations, weighing competing priorities, and making calls without a clearly correct answer. For roles where expertise is primarily technical, physical, or creative, AI’s effect on development pathways is different and, in some respects, more benign. A junior engineer using AI to write and debug code is getting more exposure to working code, faster than their predecessor did. A designer using AI to iterate concepts is building aesthetic judgment through more rapid cycles. The concern raised here is specific: when expertise depends on forming sound assessments under uncertainty, the experiences through which that capability historically developed are precisely the ones AI is most rapidly removing.
Productivity compounds quarterly. Expertise compounds over careers. AI helps organizations optimize the first. Expertise debt is what happens when they unintentionally undermine the second.
Expertise debt is a gradual mismatch between the capabilities organizations will need in the future and the experiences available to develop those capabilities today. Just as technical debt accumulates quietly until it constrains future progress, expertise debt emerges when organizations optimize for short-term productivity while unintentionally removing the developmental experiences through which judgment and professional intuition historically evolved. Whether this becomes a widespread problem remains uncertain. But it deserves serious attention, because the consequences will emerge slowly and become visible only years later.
A managing partner at a professional services firm was asked recently how she developed her judgment on client risk. She thought for a moment and said: “I spent two years doing work that, honestly, a good AI tool could do today in an afternoon. Hundreds of hours of research, first-pass analysis, drafting memos nobody read. But somewhere in those two years I developed an instinct for when something was off. When the numbers looked right but the story didn’t hold. I can’t point to the moment it happened. I just know it did.”
She paused, then added: “I’m not sure how someone develops that instinct now.”
That question, asked not with nostalgia but with genuine professional concern, is the one this article is about.
A 2025 study of 5,172 workers found AI boosted productivity by 15% on average, with the largest gains for less experienced workers, who closed much of the gap with senior colleagues. (Brynjolfsson, Li and Raymond, Stanford/MIT, QJE 2025)
That sounds like good news. In one sense it is. But the evidence on long-term effects is still emerging. AI may accelerate expertise formation in some domains by exposing junior professionals to higher-quality outputs and broader contexts than traditional apprenticeship models ever could. At the same time, research on expertise development in medicine, aviation, and military command points consistently to deliberate practice, repeated exposure to uncertainty, and learning from consequential decisions as the mechanisms through which judgment is built. The question is not whether AI helps people learn. It certainly does. The question is whether the experiences AI removes are also the experiences through which judgment historically developed.
Early career : opportunity with a catch
This generation of early-career professionals has access to tools that would have seemed extraordinary a decade ago, whether they are entering corporate roles, clinical training, research, or early-stage ventures. On paper, the development curve should be steeper than ever.
But development has never been primarily about speed of information access. It has been about the accumulation of small judgments made dozens of times a day that gradually build pattern recognition, risk instinct, and professional confidence. Three days spent on a research project was not just inefficient time. It was exposure to dead ends, conflicting sources, and the experience of a hypothesis not holding up. The hour-long AI-assisted version may produce the same output through a different set of experiences. The open question is whether those experiences cultivate the same forms of judgment, or whether something important is lost in the compression. The risk is not irrelevance. It is becoming highly capable at producing outputs while remaining relatively untested in forming assessments.
For individuals, the challenge is not to compete with AI on output but to deliberately accumulate judgment. Seek work that requires taking positions, forming assessments, and defending recommendations under pressure. Build a record not of what you produced but of the calls you made, why you made them, and what happened. That record becomes judgment capital.
For organizations, every decision about which tasks to automate is also a decision about which developmental experiences to remove. Before automating a junior role’s core activities, ask: what does this person learn by doing this work, and where will they learn it instead? Create structured debriefs where junior staff articulate the judgment calls they observed, what they would have done, and why.
Mid-career : most exposed
This is the most disrupted band and, for whatever reason, the least discussed. Mid-career professionals across industries have spent years accumulating knowledge-based expertise that formed the foundation of their professional identity. Market knowledge, regulatory familiarity, technical depth. These capabilities are not disappearing, but their scarcity, and therefore their differentiating power, may be declining.
The transition these professionals face has no obvious playbook. They are not early enough in their careers to reframe their value around the productivity boost. They are not senior enough to have built the organizational trust and credibility that makes senior expertise genuinely irreplaceable. The thing they built their career on is becoming less scarce. The thing that would replace it, judgment capital built through consequential engagement over time, cannot be acquired quickly.
For individuals, the most important shift is from knowledge depth to judgment track record. Seek work that requires forming and acting on a view, not contributing to someone else’s view but owning the assessment and living with the result. Take on one project outside your domain in the next six months that requires a recommendation without technical knowledge to fall back on. The discomfort of that position is exactly where judgment develops.
For organizations, create explicit judgment-building tracks: rotational roles, stretch assignments, positions that require consequential engagement with uncertainty. Redesign promotion criteria to reward demonstrated judgment quality, not years served. Pair mid-career professionals with senior leaders for reasoning observation, not just project work.
Senior professionals : advantage shifting, not disappearing
Senior professionals have something genuinely hard to replicate: judgment capital built through years of consequential engagement and pattern recognition across multiple cycles. That capital does not disappear as knowledge becomes more accessible. If anything, its relative value increases.
The risk worth naming is the conflation of knowledge advantage with judgment advantage. Many senior professionals have held both simultaneously. They knew more and they decided better. AI appears to be narrowing the knowledge gap faster than it narrows differences in judgment and contextual understanding. Those who cannot distinguish which was driving their value will find the transition harder than it needs to be.
The more important shift is organizational. If those earlier in their careers are increasingly relying on AI for information and analysis, the most valuable contribution a senior professional can make is no longer a better answer. It is a better reasoning process that others can learn from. The shift from expert to judgment mentor is not a demotion. It is a recognition that the scarcest thing they hold is not what they know but how they think.
For individuals, use AI to compress the information layer and create space for judgment. Make your reasoning process explicit and teachable. The next time you form a significant view, write down why, not the conclusion but the reasoning that led to it, and share it.
For organizations, shift senior performance evaluation toward two outcomes: the quality of assessments made, and the quality of judgment developed in others. Build reasoning mentorship into how senior roles are defined and how performance is assessed, not as an optional activity but as a core expectation.
Organizational leaders : strategic advantage, with an obligation
At the leadership level, the personal risk from AI is relatively low. The capabilities that define leadership value, reading organizational dynamics, navigating stakeholder complexity, making consequential calls under genuine uncertainty, are deeply human and genuinely difficult to replicate.
The organizational risk is a different matter. The pipeline producing the next generation of high-judgment talent is potentially being disrupted right now, at the early and mid-career levels, in ways that will not be visible at the top for five to ten years. Productivity metrics look fine. AI adoption numbers look impressive. The consequences may surface later, when organizations discover it is harder than expected to develop the high-judgment leaders and professionals the moment requires.
For individuals at this level, the pipeline question deserves to be treated as a structural concern with a long-time horizon, not a talent program and not an HR initiative. Before the next board meeting, ask: if we mapped the developmental experiences available to our early-career professionals today against those available five years ago, what has changed, and what have we put in its place?
For organizations, commission a talent pipeline assessment with a five-to-ten-year horizon. Make judgment development a board-level consideration tracked alongside AI productivity ROI. Start with one honest question: what is the ratio of investment in AI upskilling to investment in judgment development, and what does that ratio say about where priorities are?
None of this argues for preserving inefficient work or slowing AI adoption. Most of the work being automated should be automated. It was never the point. The point was what happened to people while inefficiency is being addressed and automation is being done at scale.
Organizations know how to measure productivity gains from AI. Far fewer know how to measure whether they are still producing expertise. The gap between those two numbers may become one of the defining management challenges of the AI era.
Part 3: If expertise debt is accumulating, how would an organization know? What are the early warning signs, and what can be done before the consequences become visible? That is the subject of Part 3.
