Why Your Data Team Shouldn't Just Use AI - It Should Live Inside It

A new report from dbt reveals a surprising gap in how data teams are adopting AI. While 72% of teams now prioritize AI-assisted coding, only 24% are focusing on AI-assisted data pipeline management. That's a big disconnect, and it points to a fundamental challenge: using AI as a personal productivity booster is easy, but embedding it into the actual infrastructure your team relies on every day is a whole different game.

The reason for the lag is simple. AI coding tools are individual - you fire up Claude Code on your own machine and go. But AI pipeline management requires team-level coordination. You need alignment on what the bots should do, how they should alert you, and who approves the changes. One client with 80 engineers touching their dbt project has only five people managing the orchestration. That's a small team trying to make big decisions, and it slows everything down. The real opportunity, though, is in layering AI on top of your existing stack without ripping it apart. Your ELT pipeline stays the same; you just add AI where it adds the most value.

Start small. Pick a low-stakes pain point like code review or error triage. Build a pilot with clear skills - write down your coding standards, your business context, and let the AI follow those instructions. Test it on a small repo, gather feedback from skeptics as well as enthusiasts, and refine. Then, and this is the part most teams skip, build evaluation loops. AI models degrade over time. If you don't check them regularly, you'll end up with noise nobody trusts. Leadership needs to hear this too: AI won't 10x your velocity overnight. It takes work to make non-deterministic tools reliable. But if you commit to the process - pilot, evaluate, maintain - you'll turn your data team from order-takers into true internal consultants.

Source: dbt Labs Blog