For thirty years we built data for humans. Cleaned, joined, normalized, dashboard-ized. A tired analyst at the other end wanted everything organized.

That’s changing. More and more, it’s agents that are reading your tables, docs, and APIs. And they see things differently.

Unorganized is OK? An agent can join messy data itself - so maybe it wants a description of how tables relate, not a pre-joined view?
Optimize differently? It pays by the token, not query - so the cheapest-to-store schema isn’t the cheapest to use?
Logs are docs? It learns more from your logs (what people actually did) than your docs (what you wished they’d do).
Code beats schema? It often works better with a function it can call than a table it has to read.

So, when data is for agents, do we engineer it differently?

I’m not fully sure. We’ll figure it out together, live, on your data.

We’ll do two things.

See how the same dataset looks to a human versus an agent - and rebuild it for the agent.
Use agents to rebuild, because they’re surprisingly good at preparing data for other agents.

If agents will read your data, or your agents will read data - join in.
That’s basically all data and AI engineers, analytics and BI folks, and anyone who publishes schemas, docs, or APIs.

It’s hands-on. Bring a laptop, a paid AI account, and datasets you have & like. You’ll leave with a checklist for: is this agent-ready?

Come do stuff. You can always watch the recording later to see stuff.

When data is for Agents, not humans

{{ gettext('Draft') }}