Waller: AI agents now make about half of all visits to the Fed's FRED database
Serguey Asael Shinder2 min readFederal Reserve Governor Christopher Waller used a speech on 1 October at FRED Con 2026, hosted by the Federal Reserve Bank of St. Louis, to describe how artificial intelligence is changing the use of FRED, the St. Louis Fed's public economic database, and to make one announcement about the Board's own data.
The traffic numbers. FRED, which Waller called the greatest public good the Federal Reserve has created, now hosts more than 850,000 data series and reached 18 million unique users in 2025. Its traffic has always grown, but since the arrival of AI the pace has picked up to about 150% a year, most of it from AI agents and bots. About half of all visits to FRED now come from AI agents retrieving data; the other half are people. It started in 1961 as a typed weekly digest mailed out by the St. Louis Fed's research director and became a website in 1995, with 865 series and about 6,000 users a week.
The problem he named. Waller said FRED's managers do not know whether AI agents attribute data correctly to the agency that produced it, or to FRED instead, and that testing has sometimes shown misattribution. He listed the familiar risks of generated content — hallucination, bias, hidden assumptions delivered with confidence, and confusing correlation with causation — and noted that the ultimate credibility of the data lies with the agencies that produce them, while FRED's role is to guarantee they are reproduced accurately. In testing, he added, agents have generally been accurate in describing charts and explaining concepts.
What FRED is doing. It has introduced the FRED MCP Connector, which uses the Model Context Protocol to let people find and retrieve FRED data through an AI agent of their choice; the staff call this "BYOAI". FRED has so far been designed for human readers, and Waller said the team now has to adapt how data are presented so that an agent passes them back accurately and with documentation. Staff are also building guidance to help visiting agents work with economic data, and links to FRASER, the archive of Federal Reserve documents, and to ALFRED, which keeps data as first published and in earlier vintages, so that AI queries can show how figures were revised.
Waller also announced that users looking for certain data published by the Federal Reserve Board will soon be routed to FRED and its tools. He did not say which series.
What it means
The speech says little about interest rates and a lot about the plumbing of economic information. For anyone who uses AI tools to pull US data, the practical point is Waller's own: check which agency produced a number and which vintage it is, because the agent may not tell you, and FRED itself cannot yet see whether it does.