In recent weeks, some of the biggest names in tech have set their sights on enterprise data, suggesting that the race to build useful artificial intelligence agents isn’t about models, but about the data those models can access and how well they understand it. Google Gemini for Work Most prominently, it shows how agents can leverage corporate profiles to perform complex tasks. Dell recently launched its own suite of tools, and Salesforce said at the Dreamforce conference that it’s eager to get in on the action.
A standard objection to using artificial intelligence to solve data problems has long been that the models aren’t good enough. I remember working on natural language queries ten years ago when Google BigQuery was being built, and the technology wasn’t ready for anything more than controlled demos. In fact, things will basically continue like this until late 2025, when advances in model and agent query patterns, as well as models that can explore materials, optimize queries, and inspect results, finally make these capabilities practical.
But making technology work is only part of the challenge. The remaining obstacles have nothing to do with the model’s ability to query the data. It’s the ability to turn data into answers that reflect how a business operates.
Why enterprise agencies need better data
If you ask an agency how many customers a company has, they need to know if a free trial counts, or if a customer who churns and comes back counts twice. It needs to know which of three slightly different revenue profiles located in three different repositories is the one the finance team actually trusts. These are not technical issues at all; They involve definitions, assumptions, and business rules that give these numbers what they mean in each specific organization.
This translation layer is often called “context”. An agent who has access to your data but lacks context is nothing more than an overconfident intern; it can do more harm than good and certainly won’t be trusted to make important decisions. The fear of artificial intelligence giving the wrong answers is one reason why these tools haven’t yet won a seat on the boardroom. Everyone worries that their CFO will confidently get some key number wrong. As usage in an organization expands beyond technical teams, risks become more dispersed. Employees across the organization can begin to rely on subtly wrong answers, allowing flawed assumptions to influence decisions before anyone realizes there’s a problem. The stronger the data and context layer, the more likely the agent is to provide answers that people can trust.
If done well, a powerful data layer can unlock tremendous value. More and more people in an organization are able to get answers to their own questions without having to wait for an expert to build a report or dashboard. This may represent the biggest shift in how people use data since the spreadsheet.
Imagine a regional sales manager wants to know which customers have gone quiet this month. Instead of submitting a ticket and waiting for someone to build a dashboard, managers can ask questions directly and get answers within seconds. We’ve seen this happen with MotherDuck and several of our more AI-focused clients. People across the company, from C-suite to sales to support and marketing, are building their own dashboards and small applications to answer questions important to their business units. Data teams no longer need to be the middleman for every KPI, report or dashboard. Instead, it can focus on making the data reliable and accessible, while employees use it to investigate problems and make decisions. As a result, people ask more questions, get more answers, and have more opportunities to make better data-driven decisions.
From data to action
This gets really exciting when you get beyond the dashboard and into the app. If a prompt can produce a dashboard that helps someone understand a problem, it may only take a few tweaks to allow an agency to build a full-fledged application that helps solve the problem. So why not use the app? Why stop showing someone what’s going on when an agent can help them decide what to do next?
Imagine a salesperson building a dashboard to understand customers who frequently have support issues. Wouldn’t the salesperson want to draft and send emails to check in on these customers? Helpful tools can combine underlying data, identify accounts that need attention, prepare personalized outreach, and give salespeople the opportunity to review before sending. When anyone can successfully build an application, we’re likely to see an explosion of small, custom tools designed around specific jobs, teams, and workflows. Collectively, they can change the way an entire organization works.
When anyone in a company can create data-rich applications, it will change the way they work. The scope of change is not limited to C-suite executives or data teams. It will extend across the organization, shortening feedback loops and speeding up decision-making. But the ability to build more tools is only valuable if those tools are based on reliable information and a shared understanding of how business works.
Models make it possible to do more with data. The extent to which companies can leverage these capabilities depends on the quality of the data and the extent to which it reflects how the business actually works. As the race to build useful AI agents intensifies, the companies most likely to benefit may be those that put in the hard work to make their data clear, trustworthy, and ready for action. So the real question is: Is your material ready?
