Anthropic released a new edition of its Economic Index, revealing a significant shift in how artificial intelligence is reshaping work as users increasingly rely on autonomous AI agents rather than traditional chatbot interactions. The report highlights that Claude usage has evolved over the past year, with long-running agentic tasks through Claude Code and Cowork becoming more prevalent. This shift has prompted Anthropic to alter its measurement of AI adoption, incorporating higher-frequency data sampling, new conversation classifiers, and the April launch of its first Economic Index Survey.

The findings indicate that AI usage closely mirrors the rhythms of daily life and work. Business-related conversations decrease on weekends, while personal uses—such as emotional support, medical questions, and investment advice—see an increase. Additionally, tax-related requests surge ahead of the U.S. filing deadline, recipe requests peak around dinner time, and sleep-related questions spike before dawn.

Anthropic also observed that Claude's outputs are becoming increasingly sophisticated. Explanations, documents, and reports are the most common deliverables. Conversations linked to higher-paying occupations generally consume more computing power, suggesting that AI is being utilized for more complex knowledge work.

Survey results show that respondents broadly expect AI capabilities to advance rapidly over the next year. More than one-third believe AI will be able to perform most or nearly all of their work tasks within 12 months, while nearly six in 10 expect AI to handle a larger share of their workload than it does today.

Despite concerns about automation, Anthropic noted that users who delegate more work to Claude are generally the most optimistic about AI's impact on their careers. Heavy AI users are more likely to anticipate improvements in pay, job security, and job opportunities. Majorities also reported productivity gains, increased learning, and greater confidence that AI is making their skills more valuable.

The company stated that these findings suggest workers continue to view AI primarily as a collaborative tool rather than a replacement. Many respondents expressed hopes that AI will automate repetitive work, allowing humans to focus on more meaningful tasks.

Anthropic has introduced Claude Tag, a new Slack-based feature that allows teams to assign tasks to Claude directly within shared channels. The tool is rolling out in beta for Claude Enterprise and Team customers, positioning it as a collaborative way to manage workflows while keeping projects moving in the background.

Claude Tag integrates into specific Slack channels, connecting to approved tools and data sources. Users can tag @Claude to initiate tasks, and the model maintains context from the channel to return results in a thread. The system functions as a shared teammate, allowing any team member to view its activity and continue requests where others left off.

An optional "ambient" mode enables Claude to proactively surface updates, flagging relevant information from connected channels and tools. Administrators retain control over channel access, tool permissions, and data visibility, with configurations designed to prevent cross-team memory sharing. Admins can also enforce usage limits and review activity logs detailing Claude's actions and request initiators.

Anthropic reports that 65% of its product team’s code is generated by an internal version of Claude Tag. Beyond engineering, teams use the system for tasks such as retrieving product numbers, managing support queues, and debugging.

Claude Tag replaces the previous Claude in Slack app, with a 30-day opt-in migration window for administrators. Eligible Enterprise and Team organizations will receive a launch credit, and the feature operates on the Opus 4.8 model. Anthropic plans to expand availability to other work environments in the future.

In a separate development, Anthropic announced results from a rerun of its "Project Fetch" robotics test. The company stated that its newer Claude models outperformed the previous generation, with Opus 4.7 completing tested objectives roughly 20 times faster than the fastest human teams.