AI Is Helping New Hires Catch Up to Veterans Faster Than Ever

Stanford economist Erik Brynjolfsson, along with MIT’s Danielle Li and Lindsey Raymond, tracked 5,179 customer support agents at a Fortune 500 software company in one of the largest field studies of generative AI in the workplace. Published through the National Bureau of Economic Research (NBER Working Paper 31161), the study found that giving agents access to a GPT-based conversational assistant increased the number of issues resolved per hour by 14% on average.

The gains were not evenly distributed. The least experienced and lowest-performing agents saw productivity jump by 34%, while the most skilled agents experienced virtually no measurable benefit. Brynjolfsson and his co-authors attribute this to a specific mechanism: the AI system had effectively learned the problem-solving patterns of the firm’s best agents and was transmitting that tacit knowledge to novices in real time. In their data, agents with just two months of AI-assisted tenure performed at the same level as unsupported agents with over six months of experience — compressing what was once a slow, intuition-building apprenticeship into weeks.

Separate research from the Wharton School reinforces this pattern, describing AI as a “leveler” that allows lower performers to make the most significant strides toward proficiency. The consistent finding across studies is that AI’s productivity benefits flow disproportionately to those at the bottom of the skill distribution.

The End of the Grunt Work Apprenticeship

Crucially, Brynjolfsson and his co-authors stop at the firm level: they conclude that generative AI boosts productivity, retention, and customer satisfaction by upskilling workers rather than replacing them. The harder question is what happens when thousands of firms make hiring and training decisions based on these gains.

If the productivity floor rises, entry-level positions become harder to justify. For CFOs, the math is becoming irresistible: why pay for three trainees when one AI-augmented junior can hit the same KPIs? Data from the World Economic Forum indicates significant pressure on entry-level job postings as companies prioritize workers who can immediately leverage AI tools over those requiring traditional training periods. By automating the foundational tasks that once served as on-the-job education, firms may be reshaping their long-term talent pipelines. The decline in U.S. entry-level postings has reached approximately 35% over 18 months, with AI as a major — though not sole — driver.

Research by Harvard and BCG involving 758 consultants reinforces this trend. Those using GPT-4 completed 12.2% more tasks and did so 25.1% faster. The gap-narrowing remained the primary finding: the lowest-performing participants experienced a 43% performance jump, while high performers saw only a 17% increase. This data suggests that AI-driven leveling is not limited to customer service but extends to high-level consultancy and analytical tasks.

A Global Pivot in the Outsourcing Capitals

The leveling of productivity is currently reshaping the global geography of labor. For decades, the economic models of major outsourcing hubs relied on selling human hours to Western companies. This model is now facing a transition. While the IMF doesn’t use “seat count” language, their data suggests that emerging markets are entering a period where they must shift from “cost and capacity” to “capability and throughput.” Instead of selling the labor of 1,000 workers to answer phones, offshore firms are increasingly marketing the quality of AI-supervised results. This is how traditional BPO models (selling hours) are pressured by AI.

According to an IMF report from January 2024, approximately 60% of jobs in advanced economies are exposed to AI, compared to 40% in emerging markets and 26% in low-income countries. However, the IMF warned in early 2026 that while emerging markets have lower immediate exposure, they risk being “left behind” if they lack the digital infrastructure to adopt these productivity-enhancing tools. The transition requires a fundamental change in how labor is valued, moving away from time-based metrics toward output-based quality.

The Gray Ceiling and the Skill Premium

The labor market is also showing signs of what can be interpreted as a “gray ceiling.” The IMF reported in early 2026 that in regions with high AI demand, employment in exposed white-collar occupations is 3.6% lower after five years. Senior workers, empowered by AI to maintain high output, are staying in their roles longer, while junior candidates find fewer available entry points. Brynjolfsson’s experiment doesn’t measure this; it’s a macro follow-on.

Yet for those who do break through, the financial rewards are significant. The PwC 2025 Global AI Jobs Barometer found that AI-skilled workers command a wage premium of up to 25%, with some specialist roles seeing premiums as high as 56%. This suggests that while AI levels the playing field for basic productivity, the market places an enormous value on those who can master the tool itself to drive complex outcomes.

Despite the rising wages for the tech-savvy, the broader labor market is adjusting. The decline in high-salary job postings for traditional roles suggests that AI is targeting analytical and expert tasks more aggressively than manual labor.

Navigating the Jagged Frontier

There is a critical limitation to the surge in junior productivity. Researchers at Harvard and BCG describe this as the “Jagged Technological Frontier.” While AI is an expert at tasks “within the frontier”—such as drafting standard legal summaries or writing routine code—it can fail significantly at tasks “outside” it.

When a task falls outside the AI’s capability, error rates can jump by 19 percentage points. This creates a dangerous paradox: junior workers now have the tools to produce high volumes of work, but they may lack the seniority and experience to verify the accuracy of the output. A “spectacular failure” in this context could manifest as a junior developer inadvertently deploying code that contains a hallucinated library, leading to a system-wide crash, or a legal researcher relying on a hallucinated precedent in a high-stakes filing.

We are moving from an era of “creation” to an era of “verification.” If a junior worker is now a “super-agent” producing the output previously expected of multiple people, the need for senior oversight becomes more acute, not less. Without senior intervention, the risk of systemic errors in AI-generated output increases, placing a higher burden on the remaining veteran staff to act as validators. Brynjolfsson’s emphasis is that AI works best when embedded in complementary processes and monitored by skilled workers, not left unchecked.

The Future of the Middle-Skill Workforce

As of mid-2026, the question for the workforce is whether this “leveling” will lead to a more equitable economy or a more crowded one. Analysis from MIT Sloan suggests that if used correctly, AI could assist in restoring the middle-skill, middle-class heart of the labor market by democratizing access to higher-value work that was previously restricted to those with years of specialized training.

However, the pressure on the modern worker is no longer just to learn a trade, but to constantly adapt. As traditional entry-level roles evolve, the new worker functions less like a student and more like a pilot—operating a powerful machine that can carry them further than ever before, provided they can identify when the system is veering off course. For the first time, the rookie in the cockpit has access to the same power as the veteran, but the ability to land the plane remains tied to the critical skill of human verification.