Anthropic’s AI Economic Index for September 2025 tracks real-world AI use. Adoption is growing fast but uneven. People at home are giving tools clear instructions for drafts and summaries. Inside firms, the wins are narrow, repeatable work like coding and admin. The big pattern is simple: if a tool does the job well, companies pay for it. Cheaper but weaker tools lose.

The bottleneck is not tokens or model choice. It is clean data, standard workflows, and clear ownership. Teams with shared data and simple processes get quick wins. Teams with messy systems stall. So an AI plan is mostly plumbing and process, not shopping for a new model.

Who moves first? Richer, better organized places and teams. Inside countries, leaders tend to have the right job mix and clear governance, not just hype. If you want faster adoption, fix data systems, clarify permissions, and train people to use the tools. Discounts help less than complements.

Two near-term bets. First, “tell the tool what to do” use keeps rising, then settles into mixed human plus automation workflows. Second, pay tilts toward people with context and judgment, while some entry tasks shrink unless managers rebuild jobs to create real on-ramps. The index is one window, but a useful read on where value shows up now.