Anthropic published two related items on 22 July 2026: a connector for its Anthropic Economic Index, and the agenda for its Economic Futures Research Fund. Neither will change your daily workflow this week, but the connector is a useful signal about the direction of assistant products generally.
What the Economic Index is
The Anthropic Economic Index is the company's ongoing research effort tracking how AI is being used across the economy — which tasks, which occupations, how usage shifts over time. It has been published as reports and data.
Making it available as a connector means the data becomes something you can interrogate conversationally rather than something you download and analyse separately.
Why a connector rather than a download
This is the interesting part. The pattern of the last two years was: vendor publishes dataset, analyst downloads it, analyst writes their own code to make sense of it. The connector pattern collapses those steps.
That is a meaningful shift in who can use research data. A policy researcher or journalist without a data-analysis background gets access to the same material as someone comfortable in a notebook. Whether the answers they get are correct is a separate question, and a real one.
The caution that comes with it
Conversational access to a dataset makes it much easier to produce a confident-sounding number and much harder to see the caveats attached to it. A downloaded dataset comes with a methodology document that you at least have to scroll past; a chat answer does not.
If you are going to cite figures from this, the discipline is the same one we recommend for any AI-mediated research: get to the underlying methodology before you publish the number, and be sceptical of any statistic that arrives without its definition attached.
Also worth noting: whose data this is
The index is built from Anthropic's own usage data, which makes it genuinely informative about how Claude is used and less authoritative about AI adoption generally. A vendor measuring its own product is a legitimate and useful source, but it is a sample with a known shape.
Treat conclusions as "how Claude users are working" rather than "how the economy is adopting AI", unless the methodology explicitly supports the broader claim.