A few years ago, I sat with a reporting team at a rail freight operator while they spent almost three weeks getting a single executive dashboard ready. Not building it, getting it ready. The actual chart-building was maybe a day’s work. Everything before that was the real job: finding the right source tables, deciding what “on-time performance” even meant across four regional systems that each defined it slightly differently, cleaning out the duplicate records, and modeling it all into something a visualization tool could sit on top of without lying to anyone. By the time it reached a screen, that dashboard had earned its keep.
Last month, I watched someone hand Claude a raw CSV export and, through nothing more than a conversation, get eighty percent of that same outcome in an afternoon. Not perfect. Not governed. But genuinely useful, and fast enough that it made me sit back in my chair a little.
So here’s the honest, slightly uncomfortable question I want to put to you: if an AI model can take a data set and generate a competent, editable, good-looking dashboard in minutes, what exactly are we still paying Power BI or Tableau, or any of the advanced visualization tools to do?
The old way: source, prepare, model, and only then present
For most of my career, the value chain looked the same regardless of industry — telecom, consulting, rail, food and beverage, it didn’t matter. Someone sourced the data. Someone else prepared it, deduplicated it, standardized it, decided what “revenue” meant this quarter. Someone modeled it into a shape a BI tool could consume efficiently. And only then did anyone touch the visualization layer, applying years of hard-earned craft about colour, hierarchy, and what a business user would actually look at first. That whole chain existed because no single tool did all of it well, and because getting from raw data to something presentable took real, specialized effort.
Prompting your way to “good enough”
What’s changed isn’t that AI got good at charts. Charts were never the hard part. What’s changed is that the friction between “I have a data set” and “I have something presentable” has mostly disappeared and it keeps getting easier. Editing is where this gets genuinely interesting. It’s not just that Claude can generate a dashboard; it’s that you can look at it and say “make the trend line monthly instead of weekly” or “I don’t trust this segment, break it out separately,” and it just happens. No ticket in a backlog. No developer’s sprint capacity to wait for. If you’re a business owner staring at a quote for a six-week BI project, that’s a legitimately hard number to defend anymore.
SMB magic, enterprise silo
But I don’t think the honest answer is “AI wins” or “Power BI wins.” I think the honest answer is: it depends on where you’re standing. If you’re one person with an Excel export trying to understand your own sales for the month, prompting your way to a dashboard is close to magic, and no advanced tool is going to compete with that speed. But the moment you’re inside an enterprise, that same convenience becomes a liability. Every person generating their own AI dashboard from their own local copy of the data is quietly building a data silo a slightly different version of “the truth,” sitting on someone’s laptop, with no governance, no audit trail, and no security review. I’ve watched “single source of truth” projects fail for exactly this reason before AI ever entered the picture; this just makes it faster to create the mess.
My instinct and I say this as someone with 26 years of bias toward doing things properly, so take it with the appropriate grain of salt is that the winning pattern for small and medium businesses isn’t “everyone prompts their own dashboard locally.” It’s building that AI-generation capability into the centralized environment the business already has: an extension of the company website, or an internal web application, sitting behind the same security and access layer as everything else. Same speed, same convenience, minus the silos.
Power BI isn’t standing still either
Microsoft isn’t standing still while this argument plays out, which is worth saying plainly. Power BI now connects to Claude, and Copilot inside Power BI is doing its own version of agent-driven, prompt-based report generation. The tool that this whole debate is framed around is busy absorbing the thing that’s supposedly disrupting it. That should tell you something about where this is actually heading it may not be AI versus Power BI at all.
The real fight might be one layer upstream
Some people I respect in this space think the real argument has already moved on from presentation entirely, toward data provisioning getting the right data, from the right source, into the right shape, safely and repeatably and that the fast-emerging Model Context Protocol ecosystem is where that fight is actually happening now, quietly, underneath all the dashboard demos.
The math nobody’s done yet
What nobody’s done yet, in any conversation I’ve had on this, is the actual cost-benefit math: subscription licensing versus custom development time versus the very real, very metered cost of AI tokens at scale. I have opinions on how that math tends to shake out, but I’d rather hear it from people living it right now than assert it from a rail yard story that’s a few years old.
So I’ll leave it open rather than tie a bow on it: if the barrier to a decent dashboard really does drop to “describe what you want,” what’s the one thing you think a BI team should be spending their newly freed-up time on instead?
