Q: What are the core steps agencies should take to prepare their datasets for AI agents? A: Agencies can follow practical priorities on how to make government data AI-ready by publishing clean, stable URLs, offering machine-readable formats like CSV, JSON or Parquet and maintaining a reliable API. They should also publish a human-friendly page with plain-English context and charts and use structured metadata such as DCAT or schema.org so agents can parse fields and units correctly. Q: Why is metadata and provenance important for trusted AI use? A: Metadata and provenance provide the context agents and users need to assess reliability, including source agency, dataset title, series ID, units, frequency, geography and methodology. Showing first-release values and later revisions, linking to archives like ALFRED and libraries like FRASER, and exposing a changelog helps prevent hallucinations and enables accurate sourcing. Q: What is an open connector and how did the St. Louis Fed use it? A: The St. Louis Fed built the FRED MCP Connector on the Model Context Protocol, an open standard that links AI systems to outside data and lets many agents fetch trusted data directly. This connector lets users retrieve FRED data through the AI agent of their choice and makes the trusted path the easiest path for agents to follow. Q: How should agencies design systems for scale and reliability as AI use grows? A: Agencies should budget for growth in storage, bandwidth and compute, use caching and content delivery networks, set fair rate limits, offer bulk downloads and monitor uptime with performance dashboards. Waller reported FRED traffic is growing about 150% a year and that about half of visits now come from AI agents, illustrating why agencies must fund these fundamentals ahead of demand. Q: How can agencies make sure AI systems credit the original data source? A: Test widely used AI tools to see if they can discover your data, read it correctly and cite your agency, and add machine-readable citation fields so agents can credit the source by default. If agents fail to credit sources, fix publishing patterns such as metadata, sitemaps and citation templates rather than blaming users. Q: What investments and governance are essential for long-term open data success? A: Open data requires steady investment and leadership: agencies should fund Chief Data Officers with real resources, maintain documentation, metadata and QA pipelines, and plan multi-year budgets for APIs, storage and security. Leaders should also consolidate access on trusted hubs when appropriate and keep users involved through advisory boards or regular forums. Q: What belongs on a minimum AI-ready checklist for government datasets? A: A minimum checklist includes machine-readable formats plus a stable API, complete metadata (source, units, methods, geography and time), versioning with revision history and timestamps, discoverability via sitemaps and schema markup, validation rules and data-quality flags, guidance like sample queries and citation templates, and clear contact information. These elements help ensure data remains trustworthy and usable when AI agents are the primary consumers. Q: How can agencies close the loop with data users, researchers and AI builders? A: Create regular feedback channels such as forums, advisory boards and public roadmaps, invite developers and AI companies to report issues and track top pain points and fix rates. Measure outcomes by asking whether answers improve in accuracy and attribution, not just whether traffic increases.