Does AI Make Workers More Productive and Who Benefits Most?

Companies around the world are spending billions equipping employees with AI assistants. Proponents claim AI will double worker output, while skeptics worry it will replace jobs or cause careless mistakes. Beyond survey speculation, recent empirical studies measuring thousands of real workers show exactly who benefits most and where AI reaches its limits.

Beyond Workplace Surveys: Measuring Real Worker Output

A landmark study published by researchers at Stanford University and MIT (NBER) tracked over 5,100 customer support agents at a Fortune 500 enterprise over several months. Instead of relying on opinion surveys, researchers measured objective metrics: resolved customer problems per hour, handling speed, customer satisfaction ratings, and employee retention rates. 

Inside the 5,100-Worker Customer Support Trial

To understand real-world impact, researchers from Stanford University and MIT (NBER) partnered with a Fortune 500 enterprise to monitor 5,100 customer support agents over several months. Rather than relying on self-reported sentiment, the team tracked hard performance metrics: resolved customer issues per hour, average handling time, customer satisfaction scores, and employee retention rates.

Across the entire enterprise, deployment of the AI assistant produced an average productivity gain of 14%. Support agents resolved queries significantly faster and encountered fewer roadblocks on tricky support tickets. However, zooming in on employee demographics revealed a far more interesting dynamic: the performance gains were wildly uneven.

The Skill Equalization Effect: Why Beginners Gain 34% While Experts Stagnate

The Stanford and MIT data exposed a profound skill equalization effect across the workforce:

• Novices and Entry-Level Staff: Saw a staggering 34% surge in productivity. With the AI system running, a customer support agent with just two months of experience consistently performed at the level of a six-month veteran.

• Top-Tier Experts: Experienced zero to minimal speed improvements, occasionally reporting that the AI suggestions felt like minor distractions.

Why did beginners gain so much ground? The AI model was trained on millions of historical chat logs from the company’s highest-performing agents. In essence, the system extracted the tacit problem-solving patterns and operational wisdom of top experts and spoon-fed those insights to newcomers in real time. For junior staff, the assistant acted as a master mentor whispering the right answer in their ear during every call.

Navigating the Jagged Technological Frontier

A companion study by Harvard Business School and Boston Consulting Group (BCG) examined how management consultants interacted with AI across various knowledge tasks, defining what they coined the Jagged Technological Frontier:

• Inside the Frontier (Creative & Structural Tasks): When consultants used AI for tasks within its core competence—such as brainstorming new product categories, drafting marketing copy, or summarizing market research—their productivity shot up by 40%, and work quality ratings improved by 24%.

• Outside the Frontier (Nuanced Logic & Data Analysis): When given tasks just beyond the model's current reasoning limits—such as evaluating complex financial data with hidden logical caveats—consultants using AI performed 19% less accurately than those relying purely on manual analysis.

The danger lies in how invisible this boundary is to the human user. AI outputs look equally confident whether the model is offering brilliant creative suggestions or hallucinating faulty logic.

The Psychological Toll: Reduced Stress vs. Cognitive Atrophy

The psychological impacts of generative AI on workers show two distinct outcomes:

On the positive side, junior agents equipped with AI reported feeling significantly less overwhelmed during customer interactions. Because the assistant handled procedural lookups and phrasing, agents experienced lower stress levels, leading to fewer escalations to managers and higher job retention.

On the flip side lies the risk of cognitive atrophy. When workers rely on AI to generate steps from their first day on the job, they miss out on the valuable trial-and-error process that builds deep domain expertise. Over time, teams risk creating a generation of operators who can execute fast with AI guidance, but lack the fundamental intuition required to resolve unprecedented edge cases when the machine fails.

Limitations and Organizational Challenges

While productivity gains in support and writing are clear, long-term questions remain. If junior workers rely on AI to generate answers from day one, will they ever develop the deep expert intuition required to solve unprecedented problems in the future?