Anthropic research finds a growing experience premium is widening the workplace AI productivity divide

The rapid integration of generative artificial intelligence into the global workforce has often been framed as a great equalizer, a tool so intuitive that it would bridge the gap between high-skilled and low-skilled workers. However, new research from Anthropic suggests a more complex and potentially more troubling reality.[1] According to the company’s second Economic Index, which tracks how its Claude AI model is being used across the economy, the benefits of artificial intelligence are not distributed instantaneously or evenly.[2] Instead, a clear "experience premium" has emerged, revealing that the longer an individual uses these tools, the more productive they become.[3] This finding suggests that AI is not a simple "plug-and-play" utility but a sophisticated skill that builds over time, a dynamic that could significantly widen the inequality gap between those who master the technology early and those who are left behind.[3][1]

The data, which analyzed millions of interactions on the Claude platform, indicates a significant learning curve that contradicts the popular narrative of "no-code" simplicity. Users who have consistently worked with the AI for six months or longer demonstrate a markedly higher proficiency than newcomers.[3][1] These "power users" don't just ask more questions; they achieve a success rate roughly 10 percent higher than those in their first month of adoption. Even after controlling for variables such as country, task complexity, and the specific model being used, experienced users maintained a distinct lead in successful outcomes.[2][1] This suggests that the value of AI is unlocked through a process of "learning-by-doing," where users develop internal frameworks for when to trust the model, how to iterate on complex problems, and how to integrate AI outputs into broader professional workflows.

This experience-based advantage is creating a new form of workplace stratification often referred to as "AI fluency."[1] As some employees develop deep expertise, they are pulling ahead of their peers in ways that traditional performance metrics may not yet fully capture.[1][4] Anthropic’s research found that high-tenure users are seven percentage points more likely to use the AI for high-value professional tasks rather than personal queries.[2] They also tend to bring more cognitively demanding work to the platform—tasks that would typically require a college degree or specialized training. This creates a "flywheel effect" where the most capable workers use AI to become even more capable, while casual or late adopters remain stuck at a basic level of proficiency.[4] The result is a widening productivity divide within organizations that could eventually harden into a significant wage and career advancement gap, essentially creating a new class of "AI-enabled" professionals.

The implications of this skills gap extend beyond individual performance to the broader corporate landscape.[4] Companies that were early to integrate AI into their core operations are now seeing a cumulative advantage that latecomers may find difficult to replicate. The index notes that as coding and technical tasks migrate toward automated API-driven workflows, the nature of human interaction with AI on consumer-facing platforms is shifting toward more collaborative, "augmentative" roles.[5] In these environments, the ability to act as an effective "AI orchestrator" becomes a primary driver of value. Managers have reported that performance disparities are emerging between employees doing identical work, driven entirely by their varying levels of AI literacy.[4] If this trend continues, the traditional labor market could see a decoupling of income and traditional experience, as "AI tenure" becomes a more accurate predictor of output than years spent in a specific industry.

On a global scale, the Economic Index highlights a persistent and growing geographic divide.[6] While usage of tools like Claude is converging within certain regions—such as across different states in the U.S.—the global gap between high-income and low-income nations remains stark.[6][5] The top 20 countries currently account for 48 percent of all per-capita AI usage, a figure that has actually increased from previous reports.[6][2] This suggests that the "digital divide" of the 20th century is being replaced by an "AI divide" in the 21st. High-income countries, which possess the requisite digital infrastructure and a higher concentration of knowledge workers, are disproportionately reaping the productivity gains of the AI revolution. Meanwhile, lower-income nations risk being sidelined, as the barriers to entry—not just in terms of software access, but in terms of the human capital required to use that software effectively—continue to rise.

Anthropic’s findings also reveal a shift in how AI is being applied as models become more integrated into daily life. Interestingly, as the general user base has grown, the average estimated hourly wage associated with tasks performed on the Claude platform has seen a slight decline, falling from $49.3 to $47.9.[2] This indicates that while power users are tackling high-value professional problems, the broader population is increasingly using AI for simpler, more personal tasks like home maintenance, weather summaries, or basic product comparisons.[2] This diversification is a sign of mass adoption, but it also underscores the risk: if the majority of the population uses AI as a convenient search engine or a "life assistant," while a small elite uses it as a high-powered cognitive engine, the economic benefits of the technology will remain concentrated at the top of the pyramid.

For the AI industry and policymakers, these results serve as a warning that access alone does not equal equity.[7] The narrative that AI will "democratize" intelligence is complicated by the reality that the technology rewards sustained, sophisticated use. To prevent a permanent hardening of social and economic inequality, there is an urgent need for institutional support in AI education and training. Organizations cannot assume that employees will naturally become proficient through exposure; they must actively foster "AI fluency" through structured development programs. Without such interventions, the "experience premium" documented by Anthropic could lead to a future where the primary driver of wealth and opportunity is not just what you know, but how long you have been practicing the art of human-AI collaboration.

Ultimately, the second Economic Index suggests that we are moving out of the "honeymoon phase" of AI adoption and into a period of structural adjustment. The data confirms that AI is a transformative economic force, but it also reveals that its benefits are conditional on human skill and persistence. As the technology continues to evolve and tasks become more complex, the gap between those who can successfully navigate these systems and those who cannot will likely grow. The challenge for the coming years will be to ensure that the learning curves identified by Anthropic are accessible to all, ensuring that the productivity boom of the AI era serves to uplift the entire global workforce rather than just a proficient few.