Anthropic says experienced Claude users iterate more carefully and delegate less

What Anthropic reported

Anthropic said on March 24, 2026 that a new Anthropic Economic Index update shows Claude users changing how they work with the model as they gain experience. In the company's phrasing, longer-term users are more likely to iterate carefully, less likely to hand Claude full autonomy, more likely to attempt higher-value tasks, and more likely to receive successful responses.

That framing matters because it shifts the story from raw adoption to usage quality. Rather than describing power users as simply doing more with Claude, Anthropic is emphasizing a more controlled and selective style of use. The official Anthropic Economic Index page, which Anthropic marked as updated on March 24, 2026, positions the project as an effort to understand AI's effects on the economy, so this latest note fits into a broader attempt to measure how real-world behavior changes over time.

Related signals from Anthropic's timeline

Anthropic shared additional context in adjacent Economic Index posts on its X timeline. The company said consumer use has become less concentrated since November 2025, with the top 10 tasks accounting for 19% of conversations, down from 24%. It also said personal queries are rising and that adoption rates across U.S. states continue to converge. Taken together, those points suggest Anthropic sees Claude use broadening beyond a narrow set of heavily repeated workflows.

Anthropic did not present this post as a headline model release. Instead, it reads like a product-and-economics signal about how people settle into steadier patterns after the novelty phase of AI use. The central claim is that experienced users are not merely delegating more. They are editing, checking, and steering more, even while moving to higher-value work.

Why it matters

For enterprise teams, the update is notable because it supports a more conservative view of agent deployment. If mature users are leaning toward tighter review loops instead of maximum autonomy, that may reinforce operational designs built around supervision, checkpoints, and iteration. For Anthropic, it also strengthens the case that the most durable AI productivity gains may come from structured collaboration with the model rather than hands-off automation alone.