The Expertise Compression: What the Research Actually Shows About AI and the Value of What You Know

Generative AI has collapsed the performance gap between novices and veterans in measurable ways — but the picture is more specific, and more useful, than the headlines suggest.

---

In a 2023 study published in Science, 453 college-educated professionals were given mid-level writing tasks — the kind of work that fills the days of marketers, analysts, grant writers, and consultants. Half were given ChatGPT. The average time taken decreased by 40%, output quality rose by 18%, and inequality between workers decreased. That last clause is the one senior professionals should sit with. The people who had spent years sharpening a craft saw their edge narrow against colleagues who had not. The study did not find that expertise became worthless. It found that a meaningful portion of what expertise produces — clean prose, competent analysis, structured argument — could now be produced by someone with less of it, faster.

This is the phenomenon quietly reshaping the careers of people in their forties, fifties, and sixties: the compression of expertise. It is neither apocalypse nor hype cycle. It is a structural change in how earned knowledge translates into market value, and the research on it is now robust enough to reason from.

The Compression Is Measurable, and It Is Uneven

The pattern that appears across multiple field experiments is consistent: AI lifts the floor more than it raises the ceiling. In the largest deployment study to date, economists Erik Brynjolfsson, Danielle Li, and Lindsey Raymond found that access to an AI assistant increased productivity among customer support agents by 14% on average, including a 34% improvement for novice and low-skilled workers but with minimal impact on experienced and highly skilled workers. The authors' interpretation was that generative AI tools function by exposing lower-skill workers to the best practices of higher-skill workers; lower-skill workers benefit because AI provides them with new solutions, whereas the best performers see little benefit from being exposed to their own patterns.

A parallel finding emerged in software. A randomized controlled study at Microsoft, Accenture, and an anonymous Fortune 100 company covering 4,867 developers found a 26% increase in completed tasks among those using an AI coding assistant — and less experienced developers had higher adoption rates and greater productivity gains.

The tools encode the top of the distribution and hand it to everyone else. If your value was sitting near the top, the market implication is uncomfortable.

For the senior professional, the implication is not that you are obsolete. It is that a significant share of what you were paid a premium to deliver — pattern recognition, synthesis of standard frameworks, competent execution of well-defined deliverables — is now accessible at markup to people earlier in their careers. Deloitte's workforce analysis put it plainly: GenAI is rewriting the traditional equation — education plus experience equals expertise — and commoditizing knowledge; when AI can instantly generate insights, draft documents, write code, or produce creative strategies, the competitive advantage no longer lies in what you know.

Where the Wall Still Stands

The compression is not total. A 2024 study of 78 workers at the derivatives trader IG Group — run by Harvard Business School's Iavor Bojinov and Edward McFowland III with Stanford collaborators — examined whether AI could help occupational outsiders match the performance of domain insiders. The answer was qualified. The findings suggest that generative AI helps everyone, including novices, more with conceptualization tasks such as generating ideas and framing problems, but the technology struggles to help novices with executing tasks, including detailed implementation and hands-on problem-solving, when they lack the necessary experience. When a task's knowledge requirements are abstract and codifiable, AI assistance goes a long way; when tasks involve concrete application and context-bound nuances, the outsider remains at a disadvantage.

The researchers call this the "GenAI Wall Effect." The Harvard Business School and Boston Consulting Group study of 758 consultants reached a similar frontier conclusion. For 18 realistic knowledge tasks within the frontier of AI capabilities, subjects using AI outperformed those without; however, for a complex managerial task selected to be outside the frontier, subjects using AI were 19% less likely to produce correct solutions compared with those without AI.

Two things follow. First, expertise still matters decisively for tasks involving judgment under ambiguity, context that cannot be easily codified, and problems where the correct framing is itself in dispute. Second, you cannot know from the outside which of your daily tasks fall inside the frontier and which fall outside. The jagged edge runs through most jobs, not between them.

What Gets Hollowed Out

The more insidious risk is not replacement but atrophy. Researchers at the University of Bath recently warned that if people begin outsourcing thinking, decision-making or interpretation to AI systems, critical forms of knowledge wither over time and create a dangerous dependency that could compromise an organisation's profitability. They distinguished between knowledge that is partially compatible with AI — rules, procedures, policies — and three forms of knowledge incompatible with AI: embodied knowledge developed through practical, hands-on experience; encultured knowledge developed through organisational culture; and embrained knowledge, which is analytical judgment and problem-solving.

A University of Toronto study cited in MIT Sloan Management Review found that using generative AI systems reduces humans' ability to think creatively, resulting in more homogeneous ideas and fewer truly innovative ones. This is not a moral argument against using the tools. It is a warning about what happens to a mind that delegates its own formation.

The career-structure implication is arguably larger. Deloitte's analysis notes that in law, GenAI can now handle much of the legal research and document review that junior lawyers typically perform, which accelerates deliv