Everyone’s racing to ship AI agents for data work. They want them to write SQL, debug pipelines, generate tests, auto-document assets and surface insights on demand. It almost feels as if the promise of self-serve analytics that data engineers have been waiting for has finally arrived.
Unfortunately, these deployments are failing simply because the agents don’t understand how the data platform actually works. They don’t know which tables to trust, whether pipelines are flaky or who owns what. They can’t trace how a schema change in…
