I just got back from Google Cloud Next in Vegas. The conversations, the energy, the people—it all pointed to something that’s been quietly happening for a while: AI is no longer experimental. It’s operational. Embedded. Boring in the best possible way.
The questions weren’t “Should we try AI?” They were “How do we scale it safely?” and “What happens if we don’t move fast enough?” When enterprise teams start asking questions like that, the shift isn’t coming—it’s already here.
I met with folks from AMEX, CVS Health, Toyota, General Mills, Cengage, and more. Different industries, same signal: AI has moved from the lab to the stack.
Sovereignty, Not Just Scale
One theme stood out: sovereign AI. Companies want AI—but with control. Not just over inputs and outputs, but the full stack: models, data, infrastructure. That’s why Google highlighted in the keynote that Gemini and Agentspace can run in private, self-hosted environments. If AI is going to be used for serious work, it can’t be a black box.
Security was the busiest part of the expo floor. The Wiz booth felt like a flagship store. Public sector teams were moving fast—agencies appointing Chief AI Officers, ditching the “government is slow” stereotype.
Ultimately, security is about trust. Anyone can build a model. Not everyone can convince a bank to run it.
What’s the Answer to “Human in the Loop”?
Agents Help Keep Humans in the Loop
AI works best when it knows what not to touch. At the conference, David Trent, CTO of Pearson, shared that 20–60% of their code is now AI-generated. But the more surprising impact came from applying AI to areas that often get overlooked: documentation, project updates, internal coordination.
Chris McAvoy, who leads Enterprise Learning and Skills at Pearson, put it this way:
“We realized we’ve switched from adding AI into our products to adding our products into AI.”
The obvious use case is code generation. The better use case is everything around it—what to build, how to share it, who needs to know. That’s where agents shine: not by replacing people, but by cleaning up the mess around them.
From BI to AI: The Second Coming of Self-Service
Over dinner with the Monte Carlo team, someone made a comparison I can’t stop thinking about: AI today looks like BI during its self-service shift. At first, data was centralized. Then tools like Looker and Tableau made it accessible. And suddenly, people were building workflows that hadn’t existed before. We’re seeing the same pattern now.
One of our clients built a custom Atolio workflow that surfaced real-time sales context for field reps. We didn’t tell them to do it—they just did it. But it worked—and solved a real problem. When users start extending the platform without asking for permission, you’re building an ecosystem.
This is how bottom-up adoption actually happens. Not through flashy demos, but through friction removal. If it's already in their flow, you don’t need to convince them to use it..
When the Future Shows Up Quietly
Late last year, after a day of meetings with a financial services client in DC, I was at the airport catching the last flight back to San Francisco. Tired, gate delayed, I spotted Thomas Kurian, CEO of Google Cloud.
I walked up, introduced myself, and we talked for a few minutes. Just a short, casual conversation—but memorable. Looking back, I imagine he was already deep in the ideas that would shape what we saw on stage at Cloud Next.
That’s what stuck with me about the whole week: the future doesn’t always announce itself with a keynote. More often, it shows up quietly—in custom workflows and repurposed tools, in hallway conversations and late-night airport gates.
Next up is TiEcon 2025 in Santa Clara and Knowledge 2025. If you’ll be at either, let me know—would love to reconnect!
