Your Junior Employees Gain 35% From AI. Your Experts? Almost Nothing.
The assumption most HR leaders carry into AI deployment is straightforward: give powerful tools to your best people and watch them get better. New research from the Federal Reserve system suggests the opposite is happening.
The 35% Reversal
Stanford and MIT researchers, whose findings are cited extensively in a March 2026 Federal Reserve Bank of Atlanta working paper, found that AI assistance raised average worker productivity by 14%. But the distribution was sharply uneven: novice and low-skill workers gained up to 35%, while high-skill experts saw minimal — and in some cases slightly negative — productivity changes.
This is not a rounding error. It is a structural finding with direct implications for how you allocate AI tools, restructure teams, and plan workforce transitions.
Why Novices Gain More Than Experts
The mechanism is intuitive once stated plainly: AI fills knowledge gaps rather than amplifying existing expertise. For a junior customer service representative, an AI copilot supplies the institutional knowledge, procedural accuracy, and response quality that would normally take years to develop. It acts as a knowledge equalizer.
For a senior expert, AI offers less incremental value. Their judgment, pattern recognition, and domain fluency already cover the terrain AI is trying to assist with. In some cases, the friction of integrating AI into an expert workflow — verification, prompt-crafting, context-switching — actually costs more than it saves.
What Executives Are Seeing
The Fed Atlanta working paper, based on a survey of approximately 750 corporate executives, confirms that labor productivity gains from AI are strengthening heading into 2026. The largest effects are concentrated in high-skill service sectors — finance, professional services, and management — precisely the domains where the novice-expert gap is widest.
As the Fed Atlanta policy blog summarizing the paper notes, companies are perceiving productivity gains that outpace what revenue figures actually confirm. This revenue realization lag — where organizations feel more productive before the financial results catch up — is a critical caveat for HR leaders building business cases around AI ROI.
The Hiring Picture Is Not What You'd Expect
A separate Federal Reserve FEDS Note on AI adoption and job-posting behavior (March 27, 2026) delivers a finding that contradicts the prevailing layoff narrative: firms actively using AI are not posting fewer jobs. AI adoption is not reducing hiring demand at the aggregate level.
Aggregate AI-driven job loss expected in 2026 remains below 0.4%, according to the Fed research. But that headline figure masks a compositional shift. Routine clerical and administrative roles are declining, while demand for skilled technical positions is increasing. The tasks AI enhances most — marketing, accounting, finance, and analytical work — are expanding. The tasks AI is most likely to replace — data entry, routine customer service, and administrative operations — are contracting.
Adoption Is Still Early
The scale of change ahead is significant precisely because adoption is still in early stages. A Federal Reserve monitoring note (April 3, 2026) found that only 18% of US firms had adopted AI by the end of 2025, with more than 20% expecting to begin using it in the first half of 2026.
On the workforce side, Gallup's 2026 data shows 50% of workers now use AI at least a few times a year, with 13% using it daily. And SHRM's State of AI in HR 2026 report finds that 92% of CHROs expect greater AI integration this year, while 39% of HR functions have already adopted AI tools.
What This Means for Your AI Deployment Strategy
The productivity distribution finding changes the calculus for workforce planning in three specific ways:
1. Sequence deployment by skill level, not seniority. If your newest employees gain the most from AI, front-loading deployment to junior and mid-level roles delivers faster measurable returns than equipping senior staff first.
2. Redefine the expert role. As AI narrows the performance gap between novice and veteran workers, the value of expertise shifts from execution to judgment, exception-handling, and oversight — roles that AI does not yet replicate well.
3. Expect a perception-reality gap. The revenue realization lag means early enthusiasm will not immediately translate to bottom-line impact. Build evaluation timelines that account for this delay rather than pulling tools when Q1 results look flat.
One important caveat: the Stanford/MIT findings that underpin the 35% novice gain were primarily drawn from customer service contexts. Whether the same magnitude applies across engineering, legal, or clinical roles remains an open question. HR leaders should treat this as directionally significant — not as a universal constant.
The productivity story of AI is not "everyone gets faster." It is "your least experienced people get dramatically faster, and your most experienced people may not." That asymmetry should be shaping your deployment roadmap today.
Sources: Federal Reserve Bank of Atlanta Working Paper (March 25, 2026), Federal Reserve Bank of Atlanta Policy Blog (March 25, 2026), Federal Reserve FEDS Note (April 3, 2026), Federal Reserve FEDS Note (March 27, 2026), Gallup (2026), SHRM State of AI in HR 2026