New AI report From Anthropic Predicts “Deskilling” Is Coming

Anthropic has released its fourth Economic Index report, revealing unexpected insights about how artificial intelligence is reshaping the workforce. While many have envisioned a future where AI handles all tasks and humans simply relax, the reality appears far more nuanced, with significant implications for workers across all industries.

The report identifies coding as the first sector experiencing true agentic automation, where AI doesn’t just answer questions but actively executes commands, accesses files, and follows through on complex plans. With tools like Claude Code and Claude Cowork now available, this automation wave is poised to spread across other industries, making the coding sector a critical indicator of what’s coming for other professions.

Central to the report is the concept of “deskilling,” a phenomenon where AI handles the most challenging aspects of a job, leaving workers with primarily simple, administrative tasks. Consider a legal secretary who once needed 17 years of experience to review legal publications and identify relevant court decisions. When AI takes over this complex analysis, the remaining work becomes basic: assembling documents, organizing files, and handling routine administrative duties. The job requires less skill, experience, and attention, potentially becoming less engaging in the process.

This contrasts with “upskilling,” where AI eliminates routine tasks and allows workers to focus on high-level responsibilities. A property manager, for example, might let AI handle bookkeeping, sales records, and market research while concentrating on negotiating contracts, securing loans, and managing stakeholder relationships. In these scenarios, the worker’s value increases as their role becomes more strategic.

The report reveals a critical limitation in current AI capabilities: task success rates decline significantly over longer durations. When humans actively monitor and course-correct AI systems, success rates remain stable. However, fully automated tasks without human oversight experience rapid performance degradation. This finding suggests that earlier projections about AI-driven productivity gains may have been overly optimistic. Anthropic adjusted its estimates, roughly halving the expected annual labor productivity growth from 1.8 to about 1 percentage point over the next decade.

Despite this adjustment, the impact remains substantial. The research shows that AI adoption is spreading 10 times faster than previous economically significant technologies from the 20th century. Notably, adoption rates are converging across different U.S. states, suggesting a more democratic distribution of AI benefits compared to earlier technological shifts where early adopters gained disproportionate advantages.

The report also highlights an important pattern: AI responses tend to match the education level of user inputs. Those who phrase requests at a higher technical level receive more sophisticated responses, suggesting that effectively managing AI systems will itself be a high-skill activity.

Looking ahead, the workplace appears headed toward a model where workers increasingly manage AI rather than perform tasks directly. However, certain bottlenecks will remain, particularly in physical, hands-on work and high-level human interactions that AI cannot replicate. These human-centric services may become relatively more valuable and expensive as productivity increases elsewhere.

The report suggests two viable career paths: embracing slower-paced, human-focused work that commands premium value, or leaning into AI-assisted productivity gains in abstract, digital industries. While concerns about widespread job displacement persist, the timeline appears more measured than some predictions suggested. The transformation is still coming, just at a more manageable pace that allows for adaptation and strategic planning.