New: Guides, the context layer for AI agentsRead the blog

START FREE
BACK TO VIDEOS
InterviewBI & VisualizationAI, ML and LLMs

BI Is Having a Midlife Crisis

2026/07/28

TL;DR: BI tools cost a lot, and they're starting to look like "just charts" — something AI can generate for free. Jacob Matson talks with Ryan Dolley, an independent BI analyst writing a Wiley book called "Building Data Apps with AI," about what AI actually replaces in business intelligence, what it doesn't, and where the work is heading.

Why BI is having a midlife crisis

Two things are squeezing BI at the same time. End users are showing up with vibe-coded HTML dashboards they built in Claude or Codex, and executives are staring at expensive renewal contracts wondering why they're still paying for a tool that "just makes charts." Charts were always the tip of the iceberg, though. The parts that actually matter — caching, the semantic layer, security, governance — were hidden from users on purpose, so now everything looks replaceable even when it isn't.

What AI actually replaces (and what it doesn't)

Ad hoc questions and small-team reporting are genuinely faster and cheaper through something like Claude connected to MotherDuck than through a traditional BI tool. What AI hasn't figured out is consistency at scale: making sure two people asking the same question get the same number, row-level security, auditability, and the OLAP-style performance a BI cache gives you.

Metrics are a social contract

Ryan's main argument is that a metric isn't just math. It's an agreement about what matters, who it matters to, and what you do when the number moves. Agents don't know any of that unless someone writes it down, so the emerging job looks less like drag-and-drop dashboarding and more like context management — capturing the business meaning an agent needs to get the answer right.

What still needs to be built

Content management for vibe-coded dashboards, a durable home for standardized metrics, and deterministic guardrails (save the SQL or Python; don't let the model recompute the number each time). Standard reports aren't going away either — AI just makes them cheaper to build, so you'll probably end up with more of them. Tools like Evidence point toward the code-based direction BI is heading.

FAQS

For ad hoc questions and small-team reporting, often yes. Asking a one-off question is faster and cheaper through something like Claude connected to your warehouse than through a traditional BI tool. Where AI falls short is consistency at scale, row-level security, auditability, and standardized reporting. The realistic split is AI for exploration and one-offs, with a governed layer underneath for the numbers everyone has to agree on.

A lot more than charts. Most BI tools include a cache or full OLAP database for fast queries, a semantic layer that defines metrics, a security model controlling who sees what, and export and dashboarding features. Users mostly interact with the charts and exports, so the tool looks like "just charts," which is exactly why executives underestimate it when AI can produce a chart instantly.

Write your metric definitions down. Don't let the model recompute them on the fly. A good definition includes the math, why the metric matters, who cares about it, and what to do when it moves — enough context that an agent can reason about it the way a person would. In practice, have the AI generate deterministic SQL or Python once, save it, and reuse it rather than recalculating inside the context window each session.

Probably not gone, but the job is changing a lot. The value of being the person who can navigate a complex BI interface is dropping fast. What's growing is understanding both the business and the data well enough to manage BI like a product — defining context and curating what agents build. Ryan's take: people overestimate short-term change and underestimate long-term change, so expect the role to evolve rather than vanish.

Yes. Standard reports like financial statements aren't going away, and there may end up being more of them because AI makes them cheaper to build. A CFO still wants to open a consistent statement every morning rather than re-prompt a chatbot for each metric. What will probably happen is AI builds the deterministic report once, then people use it as the shared source everyone references.

Related Videos

View all