The Customer Experience Layer Is Next. Are You Ready?

At Databricks Data + AI Summit this week, Satya Nadella talked about something he called the system of context. Human capital and token capital compounding together. You can’t outsource the learning, he said. Passive knowledge doesn’t build anything.

That line stayed with me because it gets at something the infrastructure announcements earlier in the week don’t fully address.

The stack finally caught up. One platform, unified governance, agents built in from the start represents real progress. But the hardest part of what’s coming is knowing what to do with it, not just the infrastructure.

Here’s what Databricks called the Infinity Campaign. Instead of building audience segments and running waterfall campaigns that are rigid and slow to adapt, you have an agent for every customer. It personalizes every interaction in real time, learns, and comes back with feedback. The campaign never ends because the customer never stops generating signal.

That’s not a new idea. It’s what the best data and marketing teams have been trying to build for years, manually, with CDPs sitting between the data and the marketing tools, reverse ETL moving data to activation platforms, and engineers keeping the pipelines from breaking. I’ve worked inside that model, and I’m building the next version of it now. The pieces work, but the seams are where things fall apart.

What Databricks announced with CustomerLake changes the assembly problem. Profile agents handle identity resolution and customer 360. Campaign agents handle activation and personalization. Data never leaves the warehouse. The reverse ETL problem becomes a governed connector. The CDP layer becomes native.

But here’s what doesn’t change. Someone still has to know what a customer 360 actually means for a given business. Someone still has to define what a good segment looks like, what a compliant signal is, what happens when the profile agent surfaces a conflict between first-party and third-party data. Someone still has to know where the model breaks down and what the downstream consequence is when it does.

Satya was right. You can’t outsource that learning. The practitioners who’ve built this by hand are the ones who know what it costs to get it right, what questions to ask before an agent runs autonomously, and what runtime visibility you need to trust the output.

More models means more things that can break. ML practitioners are more in demand with AI, not less. The platform lowers the barrier to entry. It doesn’t lower the bar for judgment.

The infrastructure is ready. The customer experience layer is next. The people who know how to build it are the ones who’ve already been inside the hard parts.