AI's Productivity Gains May Fade as Expert Pipeline Dries Up

The hidden cost of AI productivity

Generative AI is delivering measurable productivity gains across knowledge industries, with studies showing improvements of 34% for novice workers in customer service and significant speed gains for consultants. But these impressive numbers mask a structural problem: the experts extracting the most value from AI developed their judgment before these tools existed.

According to research published by the Brookings Institution, today's senior knowledge workers know what questions to ask, recognize when AI outputs are wrong, and understand what good answers look like—capabilities built through years of unaided problem-solving. The concern is that the conditions producing this expertise are eroding just as our reliance on it deepens.

Why it matters

The current productivity story assumes a steady supply of senior experts who can effectively direct AI tools. But firms are already responding to AI by cutting junior hiring by roughly 16% in AI-exposed occupations while maintaining senior staff levels. This creates a coordination failure: the juniors not hired today become the missing senior experts of 2040. We may be experiencing a productivity boom funded by spending down an inheritance of expertise we're no longer replenishing.

The expertise gap AI creates

Studies reveal a sharp divide in how AI affects different types of work. Research on customer service agents and consultants shows AI excels at information retrieval—finding existing answers and applying them to familiar problems. For routine work, AI compresses skill gaps by giving novices access to expert knowledge patterns.

But work requiring genuine expertise—designing new technologies, diagnosing atypical cases, generating novel hypotheses—operates differently. A Boston Consulting Group study found that when consultants used AI on tasks outside its capability frontier, performance dropped by 19 percentage points. The model produced confident answers; users lacking deep expertise followed them into error.

The problem is that users often cannot tell which side of this "jagged frontier" a given task falls on. Making that determination requires the very expertise that AI's productivity gains let workers bypass developing.

The developmental pathway is closing

Expertise develops through what cognitive scientists call deliberate practice: struggling with hard problems unaided, receiving feedback, and slowly building pattern recognition. A junior analyst who writes a memo with heavy AI assistance and one who works through it independently produce identical artifacts. The firm sees equivalent output, but only the second analyst is building the judgment needed to spot flawed reasoning a decade later.

Commercial AI models optimize for helpfulness, not developmental struggle. Students ask for solutions, not hints. Workers request paragraphs, not feedback. The systems deliver what users want—answers—because models that refuse to help lose users to competitors.

Recent payroll data shows firms are already responding by simply not hiring entry-level workers in AI-exposed occupations. This is rational for individual firms in the short term but catastrophic collectively. Senior expertise cannot be acquired from outside when an entire industry stops training juniors simultaneously.

A historical parallel

Early programmers worked under severe hardware constraints that forced algorithmic innovation. When hardware became cheap, the pressure to write efficient code largely disappeared from mainstream practice. The craft survived in niches, but the cognitive innovation previously forced by constraints is no longer produced at historical rates.

Similarly, when AI can brute-force acceptable solutions, the marginal incentive to develop genuinely novel approaches declines. Visible productivity rises while the invisible production of breakthrough ideas potentially slows—two effects that are not contradictory but measure different things.

The Brookings Institution analysis, authored by researchers examining the intersection of AI adoption and workforce development, argues this represents a transfer from future to present. The productivity gains we're measuring today may not survive the generation that built the expertise enabling them.

The full analysis was first reported by the Brookings Institution.