The Great Skills Leveler: How AI Compresses Experience Gaps

A landmark study analyzing 5,172 customer support agents found that generative AI increased productivity by 15% overall, but the real story lies in the distribution of those gains. Less experienced workers saw disproportionate improvements, effectively compressing the performance gap between novices and experts. This "skill compression" effect has profound implications for how organizations approach hiring, training, and talent strategy.

For decades, organizations have relied on predictable experience curves: new hires start slow, gradually develop expertise, and eventually match the productivity of senior employees. This created stable talent economics, invest in training, wait for the learning curve, reap returns from experienced workers.

Generative AI disrupts this model entirely. When AI tools enable novices to perform at levels approaching experienced workers, the economic calculus of hiring, retention, and compensation shifts dramatically. Organizations must rethink how they build teams, develop talent, and reward expertise when AI can compress what used to take years of experience into weeks of AI-assisted work.

The Research: What Was Actually Studied

This research by Brynjolfsson, Li, and Raymond (2025), published in the Quarterly Journal of Economics, analyzed 5,172 customer support agents in a controlled setting. The study measured productivity by customer issues resolved per hour and tracked performance across different experience levels.

The research design was rigorous: real customer service agents handling actual support tickets, with AI assistance provided through a conversational AI tool. The study controlled for factors like ticket complexity, customer satisfaction, and resolution quality, ensuring that productivity gains weren't simply agents rushing through cases.

The critical finding emerged from comparing performance distributions across experience levels. While the 15% overall productivity increase was notable, the real insight came from disaggregating the data. Less experienced workers saw significantly larger improvements, narrowing the performance gap with senior agents.

This "skill compression" effect suggests that AI tools disproportionately augment novices by providing access to expertise, decision patterns, and communication strategies that previously required years to develop. The research provides empirical grounding for the augmentation versus displacement debate, demonstrating that AI can enhance human capabilities rather than simply replace them.

The Skill Compression Effect: How AI Levels the Playing Field

The mechanism behind skill compression reveals why AI disproportionately benefits novices. Experienced customer support agents have internalized years of pattern recognition: common issues, effective resolution strategies, appropriate tone for different customer types, and knowledge of when to escalate.

AI tools provide novices with instant access to these patterns. When a new agent encounters an unfamiliar issue, the AI surfaces relevant knowledge base articles, suggests response templates based on similar historical tickets, and recommends resolution paths that match what top performers typically do. This democratizes expertise that previously required years to accumulate, a phenomenon with implications for how organizations structure work patterns.

Veterans still provide value, their judgment, ability to handle truly novel situations, and mentorship capabilities remain critical. But the productivity gap narrows substantially for routine work, which constitutes the majority of support interactions.

The data reveals three key insights:

Time-to-competency compressed: New hires with AI assistance reached 80% of veteran performance in weeks rather than months. Organizations can onboard talent faster and reduce the productivity drain from ramping new employees.

Experience premiums questioned: When a 6-month employee with AI performs comparably to a 3-year veteran without AI, traditional experience-based compensation models face scrutiny. Organizations must decide whether to reward tenure or outcomes.

Quality maintained: Critically, the productivity gains didn't come from rushing through tickets. Customer satisfaction scores remained stable, and resolution quality metrics didn't decline. AI-assisted novices were genuinely solving problems effectively, not just appearing productive.

Strategic Implications for Organizations

The skill compression effect forces organizations to rethink talent strategy across multiple dimensions. The most immediate impact appears in hiring and training, but the ripple effects extend to compensation, career development, and organizational design.

Hiring Strategy Shifts

Organizations should test whether AI-assisted novices can match unassisted veterans in their specific context. The customer support data is compelling, but skill compression effects vary by role, industry, and task complexity. Run controlled pilots measuring performance, quality, and customer outcomes before making wholesale hiring changes.

If skill compression proves significant, consider expanding entry-level hiring while raising the bar for mid-level roles. When AI enables rapid onboarding, organizations can hire based on aptitude and cultural fit rather than prior experience, but mid-level roles must justify their experience premiums through judgment, mentorship, and handling complexity beyond AI capabilities.

Training Program Redesign

Traditional onboarding assumes gradual skill accumulation through repetition and mentorship. AI-augmented onboarding shifts focus from memorizing procedures to developing judgment on when to trust versus override AI suggestions, a transition that requires careful consideration of how to preserve critical apprenticeship pathways.

New training programs should emphasize:

• AI collaboration skills: Understanding AI suggestions, evaluating recommendations, identifying edge cases
• Escalation judgment: Recognizing when situations exceed AI capabilities
• Quality evaluation: Assessing whether AI-generated responses truly solve customer problems
• System improvement: Providing feedback that helps train and refine AI models

Training timelines compress, if AI provides real-time guidance, extensive classroom training becomes less critical. Focus shifts to hands-on practice with AI assistance, supervised by experienced agents who model good judgment.

Compensation Model Evolution

Experience-based compensation faces pressure when AI narrows performance gaps. Organizations must decide: do we pay for years of service or for outcomes delivered?

A hybrid approach preserves incentives while acknowledging new realities. Base compensation can shift toward performance metrics, resolution quality, customer satisfaction, complex case handling, while experience premiums apply to roles requiring judgment, mentorship, and AI system improvement.

The key risk: if compensation compresses too aggressively, retention of experienced talent suffers. Veterans provide value beyond raw productivity, institutional knowledge, training capabilities, and handling truly novel situations. Compensation systems should reward these contributions explicitly. Research on China's labor market response to LLMs offers additional insights into how wage structures adapt to AI adoption.

Real-World Implementation: Two Case Studies

Large Enterprise: E-Commerce Company

A major online retailer deployed AI assistance to 2,000 customer service agents and tracked outcomes over six months. They found that agents with less than one year of experience resolved issues 18% faster with AI, while veterans saw only 6% improvement, paralleling findings from research on developer productivity with AI coding assistants.

The VP of Customer Experience used this data to redesign hiring and training. They expanded hiring of entry-level agents (reducing average salary costs), shortened initial training from 6 weeks to 3 weeks (assuming AI would provide real-time guidance), and focused advanced training on complex escalations where human expertise still dominated.

Within 12 months, overall customer satisfaction remained stable while labor costs decreased by 14%, and time-to-productivity for new hires dropped from 90 days to 45 days. The company preserved its veteran agents for quality assurance, AI training, and handling VIP customers, roles where experience still commanded clear premiums.

Small Startup: B2B SaaS Support Team

A 30-person SaaS company with a 5-person support team implemented AI-assisted support tools. Their most experienced agent, with 4 years of tenure, initially felt threatened, new hires were closing tickets nearly as fast as she was.

The Head of Support reframed the value proposition: experienced agents now focus on complex escalations, product feedback synthesis, and coaching AI systems. They adjusted compensation to reward complexity handling and mentorship rather than raw ticket volume.

The experienced agent became an "AI trainer," curating high-quality resolutions that fed into the AI's knowledge base. She identified edge cases, refined AI suggestions, and taught new hires how to evaluate AI recommendations critically.

Within six months, the team's ticket resolution capacity increased by 35% without adding headcount, and the experienced agent's satisfaction scores improved as her role shifted from repetitive work to strategic impact. The company retained institutional knowledge while leveraging AI to scale capacity. For additional perspectives on how AI assists specialized knowledge workers, see domain-specific research.

What This Means for Your Industry

While this study focused on customer support, skill compression effects likely extend to other knowledge work domains where AI can surface expertise patterns. Consider whether your industry exhibits these characteristics:

High pattern recognition value: Roles where experienced workers excel by recognizing situations and applying established solutions benefit most from AI augmentation. Examples: technical support, basic coding, document review, data entry, initial diagnostics.

Structured knowledge bases: When expertise can be codified, documented procedures, historical examples, decision frameworks, AI can democratize that knowledge effectively. Tacit knowledge that resists codification remains the domain of experienced workers.

Repetitive with variation: Tasks that repeat frequently but with slight variations (different customers, varied inputs, similar-but-not-identical scenarios) are ideal for AI assistance. Purely rote tasks get automated entirely, while truly novel work requires human expertise.

Industries likely to see skill compression: legal research, financial analysis, software development (routine coding), medical diagnosis (initial triage), content creation (standard formats), HR operations, procurement.

Industries where experience premiums persist: strategic consulting, executive leadership, creative design, complex negotiations, crisis management, highly regulated decision-making, mentorship and training roles.