I was at a conference yesterday. AI dominated every conversation — keynotes, panels, hallway talk. The energy was real.

But a pattern kept surfacing beneath the enthusiasm.

Everyone agreed: it's never been easier to ask questions of your data. What's harder to admit is what happens when you do.

Ask three people in your organization what "customer" means. Or "revenue." Or "active user." You may get three different answers. Not because people are careless. Because no one ever had to agree before.

ES-MXoQ2_IY

For years, analysts have been quietly translating between business questions and data reality. They knew which table to trust. They knew when a number needed a footnote. They held the institutional knowledge that made the data make sense.

That work was invisible. It happened before the dashboard loaded, before the report landed in your inbox. It was the human layer between raw data and a confident decision.

AI removed that buffer.

Now anyone can walk up to their data and ask a question in plain language. That's genuinely exciting. But it also means the gaps analysts were bridging — the undefined terms, the conflicting definitions, the metrics that mean different things in different departments — are now everyone's problem.

New technology doesn't always solve old problems. Sometimes it just shines a brighter light on them.

So here's the uncomfortable question worth sitting with: if your organization scaled AI access tomorrow, would the answers be trustworthy? Or would you get faster, more confident versions of the same confusion you already have?

I don't think there's a clean answer. But I think the organizations that will get the most from AI are the ones willing to ask the question honestly.

30:T33a1,

The modern data stack solved the wrong problem — and nobody noticed, because it solved it so well.

For the better part of two decades, the best minds in data engineering focused on one thing: getting data from where it lives to where it needs to be, reliably, at scale, and without breaking everything downstream. They largely succeeded. The infrastructure we built — the pipelines, the warehouses, the transformation layers — is genuinely impressive. But AI has exposed a gap we were always too busy to notice.

Clean data is not the same as understandable data.

That distinction is the defining challenge of the next era in analytics. And if you're building a data strategy today without grappling with it, you're optimizing the wrong layer.

What the modern data stack was actually built to do

Let's be honest about what the classic stack optimized for. ETL pipelines existed to move data. Cloud warehouses like Snowflake and BigQuery existed to store it cheaply and query it fast. dbt existed to transform it reliably. Orchestration tools like Airflow and Dagster existed to make sure none of those steps fell over at 3 am.

Success was measured in pipeline reliability. Query speed. Cost per terabyte. Freshness windows. The whole ecosystem was engineered around a single assumption, so baked-in that nobody thought to question it: a human would be on the other end, interpreting the results.

Dashboards were the presentation layer for human cognition. A chart doesn't need to understand what "monthly recurring revenue" means — it just needs to render a number. The human looking at it brings the context. They know which revenue lines to include, which to exclude, whether to count trials, how to handle upgrades and downgrades. The chart just draws the picture.

That assumption held for a long time. It was a

Kyle Campbell

Guides

AI and data privacy: What every business should understand (and what PowerMetrics does about It)