Lessons on AI, Data, and Talent from 30 Enterprise Executives in Seattle 

In 1964, every hotel in Seattle turned The Beatles away. It was too risky and chaotic. One hotel on the waterfront said yes. It was called the Edgewater Hotel. It is now known as the property the band famously fished out of their room window while the city lost its mind outside. 

Sixty-two years later, we gathered 30 senior executives in that same hotel for dinner and a working discussion on AI. The guest list spanned Amazon, Microsoft, Boeing, Starbucks, Zillow, Databricks, Holland America, Capital One, Walmart, and Expedia, among others. These are the people running data platforms, engineering organizations, and transformation programs inside some of the most consequential companies in the world. 

Xebia has hosted these AI-to-ROI discussions all year in cities across the United States. Not only do great relationships form, but you can also watch thinking evolve over time. And over the last two events, the thinking has moved sharply. 

Why the AI Conversation Has Shifted From Job Cuts to ROI

At the start of the year, Xebia’s AI to ROI events sparked real anxiety among attendees. The dominant question was some version of: how many jobs does this eliminate, and how fast? Leaders were being pushed to model workforce reductions before they had modeled anything else. 

That framing has largely dissolved, not because AI is slowing down. AI is getting better, quickly. It collapsed because the people actually deploying it at scale ran into two facts. 

The consensus in the room was striking: the constraint on AI value is no longer the model. It is the caliber of the humans around it, and, maybe even more urgent, the quality of the data underneath it.

What Order Should You Build AI Capabilities In?

That night, no one at the Edgewater Hotel was debating whether agents mattered. The argument has moved to sequencing. What do you build first, what do you wait on, and in what order do the pieces have to arrive so the whole thing holds together? 

When the conversation moves from "if" to "in what order," you're no longer talking about a bet. You're talking about a roadmap, and to build a roadmap, an organization needs to address the gaps.  

The sharpest gap is data maturity. It's the real dividing line between organizations that get value and those that get slideware. What surprised several leaders is where that line falls. Data maturity correlates far more with industry and the age of your data estate than with how enthusiastic your leadership is about AI. Let’s face it, excitement won’t move you up the curve. A decade of accumulated, undocumented systems can hold you at the bottom, no matter how bold the mandate. 

Most organizations have not yet decided which of the three things is their real bottleneck:

It may be all three. This year, the job for executives is to name which one dominates and decide in what order to fix it. Sequencing the wrong problem first is how good budgets produce nothing.

Why Foundational Data Talent is the Hardest Role to Fill

While the market obsesses over prompt engineers and agent builders, the executives in our room in Seattle said they are hunting for something less glamorous and more scarce: people who do foundational data work exceptionally well. 

The reason is simple. Agents act on data at machine speed. Feed them fragmented, conflicting, undocumented data, and they do not just fail; they compound errors at scale, confidently. Several leaders described the same pattern. Governance that exists on paper but not in production, no shared definitions of core business concepts, context locked in people's heads, and data quality issues discovered by stakeholders instead of alerts. Platforms built for human analysts, who could work around the mess, are now serving agents, who cannot. 

So the roles rising in demand look like this:

Excellent data, not bigger models, will decide who wins with AI. Enterprise context matters.  

What Is Xebia Axis? A Data Foundation Built with Databricks

Xebia and Databricks have tackled this problem together. It’s why we walked the room through Xebia Axis, the Agentic Data Foundation, built natively on Databricks, including Lakebase as the operational backbone for agent state, memory, and the certified-product registry. 

The idea is straightforward: give agents a foundation designed for them from the start, with people setting direction and supervising. Data contracts captured in plain language that replace months of verbal requirements gathering. A business ontology that lifts natural-language-to-SQL accuracy from roughly 60% to 90–95%. Agents that draft, test, and ship governed data products in about a day of engineering effort instead of months — with ownership, access policy, versioning, and lineage attached from day one, not bolted on later. 

The Bottom Line

The technology argument and the people argument turned out to be the same argument. So did the data argument. You can't sequence an AI program well if you don't know whether your bottleneck is broken pipelines, the wrong org design, or missing proof. And you can't fix any of them without people who understand the data underneath. 

The Edgewater Hotel said yes when everyone else said no. That is similar to the call in front of most organizations. Agents matter, that's settled. The question is order. Name your dominant constraint, build the data foundation that agents can actually stand on, and put the right people in charge of directing them. 

Want to pressure-test your own sequencing? Talk to our team about where a Xebia AI to ROI session could take your roadmap next.