The Next Shift for Data Teams: AI Demands Better Questions, Not Just Faster Answers

In a recent episode of *The View on Data*, host Faith McKenna and her colleagues spoke with Sam Ferguson, a product designer at dbt Labs, about the tangible changes AI is bringing to data work. The discussion moved beyond simple automation, focusing instead on how these tools are altering the fundamental constraints of the job.

Ferguson, who previously worked at Mode Analytics, described an early experiment with embedded natural language in SQL. The concept allowed a practitioner to write the SQL they knew and leave plain English placeholders for the rest. A key insight emerged: prompts embedded directly in the code, much like thoughtful comments, consistently yielded better results than separate chat instructions. This suggests that the discipline of writing for future human readers—explaining a filter's purpose or a variable's role—now also prepares code for AI agents.

Ferguson pointed to a historical analogy: the typist. Word processors didn't just make typists faster; they erased the constraint of expensive editing, absorbing typing into every office job. For data teams, if AI can generate countless queries or models, the pressing constraint is no longer production speed. It's human attention and discernment. The central challenge becomes identifying which questions are worth asking in the first place.

The conversation turned practical. The panel agreed that while AI can accelerate individual output, it risks creating isolation if not paired with collaboration and documentation. As one host noted, blaming 'ChatGPT' for a faulty model isn't acceptable; the practitioner who prompted and shipped it owns the result. Furthermore, the task of teaching an AI agent your company's data context—which models are reliable, your specific business logic—mirrors onboarding a new team member. It forces a necessary audit of undocumented, tribal knowledge.

The advice for leaders was straightforward: experiment quickly, but think in principles. Define what you want from AI in a workflow and establish guardrails. Use these tools to handle routine tasks, freeing up time for the complex thinking that still demands human judgment. The goal isn't to create a few AI-powered heroes, but to elevate the entire team's capability.

Source: dbt Labs Blog