The Death of the Dashboard (and What's Replacing It)

Picture the scene: a Tuesday morning, prep for a quarterly business review. The COO needs three numbers — quarterly retention by cohort, a margin breakdown by SKU, and headcount-adjusted productivity. Six months ago, this would have meant pinging an analyst, who would build (or pull) three dashboards in Tableau or Power BI, share the links, and walk through them in the meeting. Tomorrow, the COO will type those same three questions into a chat window and have answers in ninety seconds. The dashboards still exist, somewhere. Nobody opens them.

The surface story is easy to tell. The dashboard is being replaced by the chat window. The enterprise BI license, for the first time in fifteen years, is a wobbly buy. The analyst-as-dashboard-builder role is shrinking visibly. Tableau, Power BI, Looker, for two decades these were the crown jewels of the modern data stack, are quietly becoming optional.

But the story underneath the surface is bigger, and I think most of us are missing it.

Humans read dashboards. Humans run queries (well, some do). Humans interpret charts. So we optimized for human consumption: visual encoding, drill-down patterns, narrative dashboards, semantic clarity, executive summaries.

That assumption is no longer safe. The consumer of your data, increasingly, is an LLM acting on behalf of a human . And LLMs don't read dashboards. They query semantic layers, warehouses, and metric stores. They don't follow visual conventions; they parse columns and definitions. They don't ask one question at a time; they ask many in parallel. They don't get tired or stop drilling at the third level. They consume your data totally differently from how a human does, and almost nothing in your stack was designed for them.

That's the part most data leaders aren't grappling with yet. The dashboards-to-chat shift looks, on the surface, like a tool decision. It is not a tool decision. It is a fundamental redesign of who consumes your data, and the new consumer's needs are radically different from the old one.

Consider what shifts when your primary consumer is an agent rather than a human:

Governance changes. The governance frameworks most of us inherited were built for human consumers behaving in mostly-predictable ways. An agent can ask ten thousand well-formed questions in a minute, recombine answers in ways no human would, and silently infer things across boundaries no audit log will catch. Your governance layer needs to fail loudly in ways its predecessor never had to.

The semantic layer changes. Most semantic layers were built for BI consumption: clean labels for charts, drill paths for dashboards. An LLM doesn't need labels; it needs crisp, machine-readable definitions of every metric, every dimension, every join, with examples and edge cases. A semantic layer designed for humans tells you "Net Revenue." A semantic layer designed for agents tells you exactly how Net Revenue is computed, what edge cases exist, how it differs from Gross Revenue, and what you should not do with it. Most teams have the first. Almost none have the second.

Data quality requirements change. A human looking at a dashboard catches obvious oddities like a number that's clearly wrong, a chart that's clearly broken. An agent querying a warehouse will confidently summarize a wrong number into a one-paragraph answer that lands in an executive's inbox. Quality issues that used to surface visually now propagate silently. The cost of bad data goes up; the visibility of bad data goes down.

Access patterns change. A dashboard is read tens of times per day. An agentic consumer might issue thousands of queries a day, in patterns no human would generate. The cost model and query optimization assumptions baked into your warehouse were not designed for this profile.

I'll go deeper on each of these in future essays. The point I want to land here is that the death of the dashboard is not the end-state but instead is the visible symptom of a much deeper shift in who consumes your data and why. Treating it as a tool decision means missing the redesign that actually needs to happen.

The consumer transition isn't unfolding in isolation. Two other transitions are happening alongside it, equally invisibly, and the three are coupled.

The first is a workforce transition. If agents are consuming your data, then the people building, maintaining, and governing it change too. The data engineer of 2026 is, increasingly, a builder of agent tooling, not a builder of pipelines for humans. The analyst-of-2020 was paid to translate a question into SQL and a chart. The analyst-of-2026 (if that role survives at all) is paid to design the systems that let agents translate questions into SQL on behalf of everyone else. (I'll write the next essay on that.)

The second is a leadership transition. If both the consumers and the producers of your data are changing, the data leader's role evolves with them. We move from manager-of-engineers to designer-of-human-agent systems. That's not a small shift, and the people best-positioned to make it are mostly too deep in their current operating models to see it. (Essay after that.)

Three transitions, happening simultaneously, mostly under the radar of strategy decks. In my experience standing up data teams (twice, at very different organizations, in very different industries) the most expensive structural mistakes always came from optimizing one layer of the org while the layers around it were quietly shifting. I believe we're in one of those moments now, at scale.

A few patterns I see most often in conversations with peers:

The "AI chat is just a new BI tool" misdiagnosis. Treating the dashboards-to-chat shift as a vendor evaluation rather than a consumer redesign. This usually ends with AI chat layered on top of a semantic layer that wasn't built for it, unreliable answers, and the conclusion that "the technology isn't ready." The technology is mostly fine. The layer underneath it isn't.

The "let the AI/ML team own this" misdiagnosis. Punting the question to whoever is closest to the LLMs. But the consumer transition isn't an ML problem, it's a data leadership problem. Governance, semantics, quality, access, and team design are not the ML team's wheelhouse, and they shouldn't be its responsibility. If a CDO doesn't own this, no one will own it well.

The "we'll wait until it stabilizes" misdiagnosis. A position I had real sympathy for two years ago and have less of every quarter. The shift is happening to your stack whether or not you're ready for it; the only question is whether you're shaping it or being shaped by it.

I plan to write about all three transitions over the next several months and, eventually, about the contrarian take I keep hearing in private but rarely in public: that agentic AI's biggest enterprise impact won't be cost reduction (the obvious play) but revenue creation (the harder, more durable one). That essay's coming. So is the case for why most CDOs are still building for the wrong consumer, and the case for what the smaller, sharper data team of 2027 actually looks like on an org chart.

If some part of this resonated, subscribe. If it resonated and you know a peer who's quietly wrestling with their Tableau renewal right now, forward it.

— Kyle