Agents are coming for data (just slowly)
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
LLMs only became reliably capable at writing SQL in the past six to nine months, which explains why AI agents have been slow to penetrate the data space. Agents are well-suited for unglamorous data engineering tasks — detecting broken assumptions, patching schema changes, tracing pipeline failures — but struggle with business-logic decisions like how to define revenue or what counts as a customer. Proactive, unsolicited insight generation remains a research-level problem due to high false-positive rates. Three practical steps for data teams: establish context infrastructure first, write and automate evals before deploying agents near pipelines, and run on infrastructure that can handle bursty parallel query loads. Latency also matters more in agentic workflows than in human-facing ones, since query response times compound across chained agent steps.
