The Next Data Stack Has No UI: Building End-to-End Analytics with Agents and Ducks

What if your next data platform is a conversation instead of a pile of dashboards and config screens? I'm not talking about a simple agent that does SQL queries and returns answers.

Today agents can do more than that: look at raw files, design a schema, deploy infrastructure, ingest the data, write transformations, run quality checks, answer questions, and publish results and even discard infrastructure at the end. So what are we actually building here: a data agent sitting on top of an existing stack, or a data stack that an agent builds and operates itself?

In this demo talk, we'll test that distinction by asking an agent to build a working analytics stack from scratch. We start locally with a few files and DuckDB, move data and workloads to the cloud with MotherDuck when we need persistence or collaboration, and ship something useful without ever opening a traditional data platform UI.

Along the way we'll look at why the DuckDB ecosystem works well for agentic workflows and where it falls apart. We'll cover state, context, reproducibility, permissions, observability, and the gap between this demo and reality, to understand what the future looks like.

At the end of this talk, you should have a good understanding about which parts of the stack humans should still operate directly, and which parts can become an implementation detail the agent takes care of.