AI Tagged the Enterprise. Now We’re It.

There was a game most of us played when we were young, usually in a backyard, a schoolyard, or a neighborhood street where the rules were never fully explained but everyone somehow understood them. Tag. You ran until someone caught you, and then suddenly, without ceremony, the game changed. You were it.

That is what Databricks’ Data + AI Summit felt like to me. Not because AI has suddenly arrived. It has been arriving loudly for some time. But because the conversation around it has changed. The technology is no longer waiting outside the enterprise, asking to be admired. It has crossed the threshold. It is inside the room now, and the responsibility has shifted.

The Summit brought together the expected mix of builders, buyers, brands, operators, partners, executives, and technical teams. Databricks said more than 30,000 people were expected in San Francisco for the event, with 800-plus sessions spanning data engineering, warehousing, governance, analytics, applications, agents, and AI. But the value of a room like that is not only in what gets announced from a stage. It is in what becomes visible when thousands of people who are trying to make AI useful inside real companies start circling the same set of questions. What can we trust? What can we govern? What can we afford? What can we automate? What still needs human judgment? And perhaps most importantly, what happens when the demo is over and the system has to work on Monday morning?

For the past year, the public conversation around AI has been driven by capability. A model can write, code, summarize, search, analyze, generate, recommend, and respond in ways that would have sounded impossible not long ago. That stage mattered because it expanded the imagination of what work could become.

But at Databricks, the more interesting conversation was not whether AI could perform. It was whether AI could hold up inside the real conditions of a company. That is a harder test. Inside an organization, intelligence has to move through budgets, legacy systems, security reviews, customer expectations, brand standards, compliance questions, internal politics, and the habits of how people actually get work done. It has to be useful to people who do not speak in technical language and dependable enough for leaders who cannot afford to confuse speed with progress.

The day one keynote made the shift concrete. Databricks CEO Ali Ghodsi spent almost none of it arguing that AI is capable; he treated that as settled, even telling the room that by any definition he’d have accepted years ago, AGI has already arrived. “AI does not have an intelligence problem,” he said. It has a context problem. The models are smart enough. They just don’t know anything about your company.

This is where the Summit’s announcements started to connect. The language was often technical — agents, real-time data, governance, security, customer intelligence, cost controls — but the underlying question was practical: what does AI need around it before a company can responsibly let it matter?

One answer was business context. Genie One, Databricks’ agentic coworker for business teams, points toward a future where AI is not just something technical teams manage from the side but something finance, marketing, operations, and sales use in the daily rhythm of work. That sounds simple until you think about what it requires. The system has to understand not just the data but the business meaning around it. AI does not simply plug into a company; it asks the company to explain itself. What does this metric mean? Which source should be trusted? What happens when the answer is fast, confident, and not quite right?

Databricks answered that last question with a demo built around failure. Ken Wong, a senior director of product management, ran one real prompt, profile the customers on our advisory board, against five leading AI agents. One returned a fast, clean answer: 24 customers. It was wrong, and when Wong asked where the number came from, the tool, in his words, “confessed that it just completely made it up.” Others came back stale, or burned half an hour and a pile of tokens and still finished incomplete. The point was not that the tools were bad. It was that a confident answer and a correct one are not the same thing, and the difference is whether the system is grounded in a company’s live data or just guessing at it.

Those same demands, grounding, freshness, trust, showed up across the Summit’s announcements. Lakehouse//RT and the broader push toward real-time analytics raised a simple but important point: if AI is going to help companies act, it cannot be working from stale or partial information. CustomerLake brought the conversation closer to brands and marketers, where AI is not just a productivity tool but part of how a company understands and responds to customers. The Panther acquisition and Security Lakehouse story underscored the same reality from another angle: once AI starts moving through the business, trust and security cannot be bolted on later.

That last point got its own announcement. Databricks said it’s acquiring Panther, a security company, and folding it into what the company calls a security lakehouse, with an agent layer it named Lakewatch that triages alerts and hunts threats automatically. Ghodsi’s argument for why was the part worth hearing: the same AI advances making agents useful are also arming attackers, so the volume of security data a company has to watch is exploding at the exact moment it can least afford to filter it down. Putting security on the same governed foundation as everything else, rather than in a separate, expensive system off to the side, was his answer.

The brand presence mattered because it made the conversation feel less abstract. AI is no longer sitting at the edge of the organization as an innovation project. It is moving into customer experience, marketing, analytics, security, finance, operations, and product. Once that happens, the conversation naturally becomes less about novelty and more about accountability. Someone has to know what the system is acting on. Someone has to understand what it costs. Someone has to decide whether the output is good enough. Someone has to protect the customer and the brand while the company learns to move faster. And somewhere, inevitably, someone is still reconciling three versions of a spreadsheet while being told an agent will solve the workflow by Q4.

That is not a knock on the technology. It is a reminder of the environment the technology is entering. Companies are living systems. They are full of workarounds, exceptions, inherited processes, institutional memory, and human judgment that rarely fits neatly into a product diagram. The next phase of AI will be shaped by how well companies can bring those realities into the system instead of pretending they do not exist.

That is why one of the most interesting stories from the Summit may not be the biggest keynote moment. It may be a session like Virgin Atlantic’s “What Worked, What Didn’t,” which is a title that already tells you something useful. The company described a generative BI effort that began with a familiar promise: faster answers for leaders who needed to make decisions across commercial, operational, and customer data. But faster answers did not automatically create more confidence. According to the session description, Virgin Atlantic’s first attempt “didn’t go well” because answers arrived faster, but confidence dropped when shared metric definitions, visible lineage, and governance were not designed in from the start. Ambiguity scaled faster than trust. That is the grown-up AI story. Not that the technology failed. That would be too simple. The lesson is that intelligence without trust can make a company move faster in the wrong direction, or at least make everyone ask the same question again: is this right?

A similarly human thread showed up in 7-Eleven’s session on AI agents for frontline maintenance. It is not the kind of example that usually gets the most attention, which is exactly why it matters. The problem was not glamorous. Technicians needed answers across PDFs, spreadsheets, shared drives, equipment images, parts information, and scattered documentation. The session described a GenAI agentic system using Mosaic AI, Agent Bricks, and Databricks Hybrid Search to help technicians retrieve maintenance documents, capture equipment images for troubleshooting, access parts information, and search for solutions. That is where AI becomes more interesting to me. Not as a spectacle, but as a way to move knowledge closer to the moment where someone needs it.

There was also a quieter but important thread around cost and control. As companies experiment with agents, usage can grow quickly and invisibly. Axios reported that Databricks introduced Unity AI Gateway after seeing customers accidentally run up AI bills into the tens of millions in a single month, with the product aimed at spend limits, runaway-spend protections, and cost management across providers. The excitement of “let’s see what this can do” eventually runs into the less glamorous question of who is paying for all those tokens, workflows, and experiments. This is where governance becomes less like bureaucracy and more like adult supervision.

The Summit seemed to capture that exact transition. AI is still exciting. It is still moving quickly. It is still opening up real possibilities for companies and brands. But the conversation is becoming more serious because the stakes are becoming more real. The demo phase is not over, but it is no longer enough. A demo can impress with a single answer. A business needs the answer to hold up across teams, systems, customers, costs, and consequences. A demo can make AI look effortless. A company has to decide where the effort goes, who owns it, and what happens when the system changes the way work gets done.

That is the harder chapter of AI. The technology is becoming more capable, but it is also asking more of the organizations that use it. Better context. Better accountability. Better design. Better judgment. Better leadership. The companies that succeed will not simply be the ones that adopt AI everywhere. They will be the ones that understand where it belongs, what it should improve, and what needs to remain meaningfully human.

By the end, the Summit’s real message felt less like “AI is here” and more like “AI is asking more of us now.” That may be the part worth paying attention to. Databricks gathered the people who are building the tools, buying the platforms, protecting the systems, shaping the brands, and trying to turn possibility into something durable. What came through was not only excitement, but the strain of making a powerful technology useful without letting it outrun the business it is supposed to serve.

AI tagged the enterprise. Now we are it.