Data Agents Will Replace Dashboards: But Who Builds Them?

P3 Adaptive’s Rob Collie asks the question: Data agents will replace dashboards, but who will build them? This article originally appeared in Insight Jam, an enterprise IT community that enables human conversation on AI.

If you follow the business intelligence (BI) industry, you’ve watched the dashboards-to-agents conversation build for a year or longer. And almost all of it has been a shopping conversation. Which semantic layer. Which large language model (LLM). Which agent framework. Reasonable questions, every one – but they’re all versions of: what should we buy?

Almost nobody is asking the question that actually decides the fate of these projects: who builds the data agents?

Our company has spent the past year building data agents for large enterprises, so I’ve had a front-row seat to that question. But what keeps striking me is how familiar it feels. I spent more than 20 years in the BI world before this – first at Microsoft as a product leader on Excel and as a founding engineer on what became Power BI, then as CEO of a BI implementation company – and the parallels are hard to ignore.

We now take it for granted that BI lives close to the business. It wasn’t always that way – for years it was the subject of intense debate and resistance. But market forces and ever-more-accessible tools wore down even the most stubborn holdouts, and BI tools are now associated more with knowledge workers than with information technology (IT).

I believe we’re about to watch the same story play out with data agents. Today they’re perceived as an IT mission, and there are valid reasons for that – just as there were valid reasons for BI to begin as one. But data agents will follow the same arc. The mission will again migrate toward the business.

Why we’re starting in the same place

BI began as an IT mission primarily because the early tools were far too technical for business users. If you hadn’t spent your career developing comfort with developer-style tools and abstract languages (as IT professionals had), the early BI platforms were unapproachable. The business stayed on the sidelines and stuck with the tool they understood: spreadsheets.

Today we find ourselves in a similar place with AI. The business is becoming comfortable with off-the-shelf subscription tools like ChatGPT, Claude, and Copilot, but that’s a far cry from building customized agentic systems.

So it makes sense that data agents are starting out centralized under IT. But success will rest on how quickly organizations remember the lessons we learned in BI, and this time we won’t have a decade to figure it out.

Why the mission will shift

Data agent success will require hands-on involvement from the business for the same reasons BI ultimately did: the complexities of the business are too much for IT to keep up with.

In the same way that early BI tools took a career to learn, the nuances of business operations require a career’s worth of experience to absorb. It is unrealistic to expect IT to digest all that nuance and reach a level of comprehension approaching that of a mid-level operations manager with two decades of experience – and even more so when you consider that IT is always outnumbered. For every IT project manager, there are dozens of business stakeholders, spanning many distinct specializations – and again, each powered by a career’s worth of experience.

Put bluntly: IT was never going to sufficiently understand business requirements. You can have as many meetings as you want and write requirements documents to exhaustion. And you will still miss the mark. This is what always happened with IT-centric BI. Always. Seriously, it never worked. Each project just reached the point where everyone’s best option was to call it a success and move on.

And even compared to BI, I believe data agents are more dependent on – and sensitive to – business nuance. By design, they will be used more frequently and more intensively than dashboards ever were. Similarly, they will be used by a broader and less sophisticated audience than BI. And raising the stakes significantly, they will also be increasingly empowered to exercise judgment and take action. For these reasons, data agents (and indeed, most agentic AI) will succeed or fail based more on business understanding than on technical acumen.

The one in 16: who absorbs the shift

As more of the BI workload migrated to the business, it did not land on random people. It instead landed on a very specific subpopulation within the business. Roughly one in 16 knowledge workers, in my experience, is born with an innate itch to solve problems with tools. They were the Excel power users. Then they became the Tableau and Power BI generation. They were the ones who finally broke the BI bottleneck – by expanding the labor pool for building BI systems beyond IT’s limited staffing, yes, but also by having the required business nuance “pre-installed.”

I call these people the Crafters, and once again they will be pressed into service to make data agents work for the business (and to help with agentic AI of all forms). Crafters are utility players who are close to – and familiar with – the business, but who also possess a talent and interest for systems thinking.

Systems thinking is the skill which differentiates Crafters from the other 15 out of 16 business workers. They see problems and immediately start breaking them down into steps which can be automated, chained together, and logically controlled. This is a skill we associate more often with software developers, and while developers do have an edge on Crafters in this skill, Crafters’ capabilities are more than a match for many business technology problems. And Crafters’ proximity to the business gives them an inside track – command of business nuance – which very few developers can claim.

No, they’re not “citizen developers”

The industry already has a name adjacent to this idea, and it’s the wrong one. The citizen developer pitch says that with the right program and the right governance, roughly 40 percent of your workforce can be enabled to build. My number, drawn from user studies I conducted at Microsoft and then again with clients, is closer to 6 percent. That’s not a counting dispute – it’s a definitional one. The 40 percent counts people who can be trained to use a tool. The 6 percent counts people who will go build something nobody asked them for.

No governance program ever created a Crafter, any more than one ever created a spreadsheet guru. The strategic consequence: your job is to find these people, not to manufacture them. That single reframe will save you a great deal of enablement budget.

How to start finding them

First, inventory the business-created systems that are both mission critical and clever. The spreadsheets and dashboards. The automations in tools like Airtable, Zapier, or Power Automate. The SharePoint apps. Then go looking for their makers. That’s your starting talent pool.

Making them effective builders of AI systems

Finding your Crafters is the fast part. Making them effective takes three investments – two from them, one from you.

Their first investment: learning how AI systems actually work. Not how to train a model – they’ll never touch the model – but how an agent’s competence gets assembled around the model: which data it can see, which tools it can use, and what context it’s handed at the right moment.

Their second: getting comfortable with coding agents – AI that writes and runs software on their behalf – because Crafters skilled with coding agents can develop real software for the first time. They arrive with business experience and technical problem-solving mindsets; these two added skills launch them into a new era.

Your investment is easier, because the platform vendors are already making it for you: sandboxed, low-code frameworks are emerging that give Crafters a place to build real systems without putting production data at risk.

The head start

The platform question will settle itself – the vendors, as we just saw, are highly motivated to settle it, and their offerings won’t differ enough to be decisive. The staffing question is more critical, and nobody is going to sell you the answer. But the BI era left us one piece of good news: the beginnings of an answer are already on your payroll. BI’s migration to the business took more than a decade. Data agents will be understood as a business-centric mission much quicker – perhaps as soon as this time next year – and the head start goes to whoever engages their Crafters first.