Google, Dell and Salesforce Bet on Enterprise A.I. Agents
In recent weeks, some of the biggest names in tech have set their sights on enterprise data, an implicit admission that the race to build useful A.I. agents is not about the model, it’s about the data those models can access and how well they understand it. Google’s Gemini for Work is the most prominent, demonstrating how agents can leverage enterprise data to perform complex tasks. Dell has recently rolled out its own suite of tools, while Salesforce signaled at its Dreamforce conference that it, too, is eager to get in on the act.
For a long time, the standard objection to using A.I. for data problems was that the models just weren’t good enough. I remember working on natural language querying a decade ago while building Google BigQuery, and outside of controlled demos, the technology just wasn’t ready. That’s really where things largely stayed until suddenly, late in 2025, when advances in models and agentic query patterns, models that can explore the data, refine their queries and check their own results, finally made these capabilities practical.
But making the technology work is only part of the challenge. The remaining blocker is not about the model’s ability to query data. It’s the ability to translate that data into answers that reflect how a business operates.
Why enterprise agents need better data
If you ask an agent how many customers a company has, it will need to know whether free trials count, or whether a customer who churned and came back counts twice. It needs to know which of the three slightly different revenue figures, sitting in three different databases, is the one that the finance team actually trusts. These aren’t technology questions at all; they’re about the definitions, assumptions and business rules that give those numbers meaning at each specific organization.
This translation layer is typically referred to as “context.” An agent with access to your data but insufficient context becomes little more than an overconfident intern; it can do more harm than good, and certainly isn’t going to be trusted with important decisions. The fear of A.I. getting answers wrong is one reason these tools have yet to earn a place in the boardroom. Everyone is worried that their CFO will get some key number confidently wrong. As usage spreads in an organization beyond technical teams, the risk becomes more diffuse. Employees throughout an organization may begin relying on subtly incorrect answers, allowing flawed assumptions to influence decisions before anyone realizes there is a problem. The stronger the data and context layers, the more likely agents are to produce answers that people can trust.
When done well, a strong data layer unlocks enormous value. More and more people across organizations are able to get answers to their own questions without having to wait for a specialist to build a report or dashboard. This could represent the biggest realignment in how people use data since the spreadsheet.
Imagine a regional sales manager who wants to know which accounts have gone quiet this month. Instead of filing a ticket and waiting for someone to build a dashboard, the manager can just ask the question directly, and get an answer back in seconds. We have seen this play out inside MotherDuck and at several of our more A.I.-forward customers. People across a company, from the C-suite to sales to support to marketing, are building their own dashboards and small apps to answer the questions that matter for their slice of the business. The data team no longer has to serve as the go-between for every KPI, report or dashboard. Instead, it can focus on making 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
Where this gets really exciting is when you get beyond dashboards and into applications. If a prompt can generate a dashboard that helps someone understand a problem, it may take only a minor tweak to get the agent to build a full-blown application that helps solve it. So why not go with the app? Why stop at showing someone what is happening when an agent can help them decide what to do next?
Imagine a salesperson who builds a dashboard to understand customers experiencing recurring support issues. Wouldn’t that salesperson also want to draft and send emails to check in on those customers? A useful tool might combine the underlying data, identify the accounts that need attention, prepare personalized outreach and give the salesperson a chance to review it before sending. When anyone can build applications successfully, we’re likely to see an explosion in small, custom tools designed around specific jobs, teams and workflows. Collectively, they could change how work gets done across an organization.
When anyone in a company can create data-rich applications, that’s going to change how they do their jobs. The scope of the change isn’t limited to executives or the data team. It will extend throughout the organization, shortening feedback loops and speeding up decisions. But the ability to build more tools is valuable only if those tools are grounded in reliable information and a shared understanding of how the business works.
Models have made it possible to do more with data. How far companies can take these capabilities depends on the quality of their data and how well it reflects the way the business actually operates. As the competition to build useful A.I. agents intensifies, the companies best positioned to benefit may be those that have done the harder work of making their data legible, trustworthy and ready to act on. The real question, then: Is your data ready?
