Build a Dashboard in 20 Minutes - The Hard Part Remains

The twenty-minute build is real. The trust, the testing and the ownership are not faster, and that is where the work now lives.

Building the dashboard was never the slow part, and AI making it faster does not change that. The work that decides whether anyone trusts it still sits with a person.

This connects with our AI work and, separately, with AI Does Not Fix Bad Data. It Amplifies It.

The demo is impressive, and that is the problem. Point an agent at a semantic model, describe what you want, and twenty minutes later you have a working dashboard. It queries real data. It looks polished. It would have taken a person a day or more to build by hand. The natural reaction is to assume the job just got twenty times faster.

It did not. The build was never the bottleneck. What takes the time, and what matters, is everything that turns a dashboard into something a business will act on. AI has made the cheap part cheaper and left the expensive part exactly where it was.

Two kinds of autonomy.

It helps to separate two things that get blurred together. There is the autonomy to build something, and there is the authority to decide whether it should exist and whether its answer is right. An agent now has plenty of the first. It has none of the second.

A model can generate a dashboard. It cannot tell you that the margin figure is using last year's cost base, or that the regional split quietly excludes a business unit, or that the number on the screen will be read in a board meeting and acted on. It does not carry the consequence. You do. The agent does not get the call when the figure is wrong in front of the leadership team.

Generate a dashboard in twenty minutes and you have produced an answer. You have not yet decided whether it is the right one.

Twenty times the output, twenty times the everything.

When the build gets faster, the work does not disappear. It moves downstream and multiplies. Build at speed and you also produce, at speed, the things that follow a build: checking the numbers, testing the edge cases, keeping visuals consistent, and earning the trust of the people who will use it. None of that got faster. If anything it got harder, because there is now more output flowing into it.

There is a specific failure mode worth naming. AI-built work is confident. It looks finished. A dashboard generated in twenty minutes presents itself with exactly the same polish whether the logic underneath is sound or quietly wrong. The faster you generate, the faster you generate convincing mistakes, and the convincing ones are the expensive ones, because they are the ones that slip through.

The question that matters.

So the speed is real and it is useful. We use these tools. But the value of a dashboard was never in how fast it was assembled. It was in whether the right person can stand behind the number. Before you celebrate the twenty-minute build, answer a plainer question: who owns whether this should exist, and who is checking that it is right?

If the answer is "the agent built it and it looked fine", you have not saved time. You have moved the risk somewhere you are not looking.

A small story, repeated everywhere.

The pattern plays out the same way across teams. Someone generates a dashboard in an afternoon, shares it, and it looks finished, so people start using it. A fortnight later a number looks off in a meeting. It turns out a measure was averaging where it should have summed, or a filter was excluding a region nobody noticed. The dashboard was never wrong in a way you could see. It was wrong in a way you had to check for, and nobody did, because it arrived looking complete. The speed that built it was also the speed that skipped the verification.

This is not an argument against the tools. It is an argument for spending the time they save in the right place.

Where the saved time should go.

If a build that took a day now takes twenty minutes, you have not lost the day. You have freed it. The question is where it goes. Spent well, it goes into the work that was always undersized: checking the logic, testing the awkward cases, talking to the people who will use the thing about whether it answers their real question, and getting the design clear enough that the answer is hard to misread. That is the work that turns a generated dashboard into a trusted one. Spent badly, the saved time goes into generating three more dashboards nobody checked, which is not progress. It is just more surface area for the next wrong number to hide in.

What we would do this quarter.

Use the speed. Let agents take the assembly off your plate, because that frees people for the part that needs them. But pair every fast build with a named owner who checks the logic and signs off the output before anyone relies on it. The build is the cheap part now. Spend the time you saved on the part that was always the point.

If any of this sounds familiar, talk to us about your data.

You can build a dashboard in twenty minutes. That is the easy part.