The Skills That Matter When AI Tools Do the Building

When anyone can generate a dashboard, knowing the tool stops being the edge. The skills worth building once the building is cheap.

When anyone can generate a dashboard, the value moves to the things the tools cannot do for you: design judgement, governance, and owning the decision. These are the skills worth building now.

For more on this, see our AI work. It is also worth reading alongside What an AI analytics project costs, and how to budget for one if the topic is new to you.

For a long time, a lot of a data professional's value sat in knowing the tool. Where the settings were, how to coax a visual into behaving, which workaround unlocked which requirement. That knowledge was scarce, so it was worth paying for. It is becoming less scarce by the month. When an agent can generate a working dashboard from a sentence, knowing where the buttons are stops being the differentiator. So this is a fair question to ask of yourself and your team: if the build is getting cheap, where does the value go?

It goes to the things the tools still cannot do for you. Here is where we would invest.

Judgement about what to build.

The hardest part of reporting was never the building. It was deciding what deserved to be built, for whom, and what to leave out. An agent will happily build whatever you ask, including the wrong thing, beautifully. The person who can look at a request and see the real decision underneath it, name the audience, and cut the noise is worth more than ever, because that judgement is now the scarce input. The tool supplies the output for free.

Design discipline.

Full control of the canvas sounds like a gift, and it is also a trap. When nothing stops you, restraint has to come from you. The ability to make a complex thing simple, to choose the one chart that makes a decision clear instead of the five that show off, is a craft, and it gets more valuable as the constraints that used to enforce simplicity fall away. Code-first reporting in particular rewards mature, structured design practice and punishes its absence.

When the tool can do anything, the question stops being "can we" and becomes "should we". That question has no button.

Governance and ownership.

We have written about gates and sign-off and context. The skill behind all of it is being willing to own the decision: to say that a number is right, or that a dashboard should not exist, and to stand behind it. An agent never owns the consequence. A person always does. The people who can carry that responsibility, and build the light processes that support it, are the ones who make AI safe to use at speed. That is a skill, and it is learnable, and it is in short supply.

Knowing the model and the meaning.

Tools churn. The semantic model and the business context underneath it do not. Understanding how the business really works, how its metrics are defined and why, and being able to encode that clearly, is durable in a way that tool knowledge is not. It is also the thing that makes every AI-assisted build better, which makes the people who hold it more valuable, not less, as AI spreads.

What gets less valuable.

It is worth being honest about the other side. Time spent memorising menus, mastering fiddly workarounds, or being the only person who knows how a fragile report holds together: that is worth less than it was, and it will keep falling. None of it is wasted, the experience underneath it is exactly what feeds good judgement, but it is no longer where the value sits. Pointing your development at it is pointing at the past.

What this means for hiring.

If judgement is the scarce input, it changes what you look for when you build a team. The most useful question in an interview is no longer "can you build this", because increasingly the tools can. It is closer to "here is a request, what would you push back on". You are hiring for the instinct to find the real decision underneath a vague ask, the restraint to leave things out, and the willingness to own an answer. Those are harder to assess and far more valuable than tool fluency, which can now be picked up in weeks.

It also changes how you develop the people you already have. A capable analyst who has spent years mastering the tool has exactly the raw material that good judgement is made of. The move is to give them problems that stretch the judgement rather than the tooling: ambiguous requests, design trade-offs, sign-off responsibility. The experience is not wasted. It is the foundation. It just needs pointing at the part of the work that is becoming scarce.

A note for leaders.

For anyone running a data function, the risk is investing in the wrong scarcity. Another round of tool training feels productive and is easy to buy. It is also the thing depreciating fastest. The harder, more valuable investment is in judgement, design and ownership: slower to build, impossible to download, and the real source of advantage once the building is cheap.

What we would do this quarter.

Pick one judgement skill and develop it on purpose. For an analyst, that might be learning to interrogate a request until the real decision shows up. For a team lead, it might be designing the sign-off process that lets people trust AI-built work. Choose something the tools cannot do for you, and put your effort there. That is where your value is heading, so it is where your attention should go.

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

The skills that matter when the tools stop saying no