Christoph Casati | 12/22/2025
On a cold December morning in Washington, DC, I walked into the Databricks World Tour expecting the usual technology conference fare: vendor pitches, product demos, buzzwords about AI transformation. What I witnessed instead was something more concrete and unsettling to traditional IT models—government agencies and international organizations describing how their business users are now building production AI systems in weeks, not waiting months for IT to build them.
The crowd was massive. Conference organizers had tripled capacity and still turned people away. In the overflow room and hallways, government employees waited to hear how their peers were breaking free from what Rory Patterson, Chairman of Databricks Federal, called “the 20-year technology gap.”
This isn’t a story about artificial intelligence replacing people or about cutting-edge research. This is about something more fundamental happening in real organizations right now: the people who understand the mission—fraud investigators, treasury analysts, investment officers—are building their own AI tools. And they’re doing it on production systems, managing real money and making real decisions.
The AI Transformation Nobody Saw Coming
AI has changed faster than anyone predicted. Two years ago, large language models were research curiosities. Today, nearly everyone in the conference hall raised their hand when Patterson asked who uses AI daily in their personal life. They use it to write emails, research medical questions, prepare for interviews, even write performance reviews.
“I wrote all of my team’s performance evaluations last quarter,” Patterson admitted to laughs. “Actually, it was the best work I’ve ever done. It gave me clarity on what I was already thinking.”
But he pointed out the obvious problem: people use AI constantly at home, then come to work and face systems that feel 10 to 25 years behind. One government agency employee I spoke with during the break said their office still uses software from 2006. The disconnect is stark.
Databricks saw this gap and made a bet: integrate AI directly into their data platform. Not as a separate product to buy, but built into the tools people already use. Their AI/BI product called Genie lets users ask questions in plain English and get charts, graphs, and analysis back. No SQL required. It’s grown 500% in the past year, and it’s free for Databricks customers.
But Genie is just the starting point. The bigger transformation is about letting organizations build their own AI agents—automated systems that can query databases, read documents, search the web, and provide insights. Databricks calls this capability AgentBricks, and it’s the reason business users are suddenly building AI instead of waiting for IT.
The Old Way: Crystal Reports and Month-Long Waits
Kapil, a treasury manager at IDB Invest (the Inter-American Development Bank’s private sector arm), stood on stage and described what many in the audience recognized from their own experience.
“I have never seen this in the last 20 years of my career,” he said. “We moved from the time where we had Crystal Reports. If you remember Crystal Reports, anything you need, you need to go to your IT, who has to work, create a report, and the turnaround time was one month or so.”
One month to get a report. Then if you need it changed? Start over.
Igor Valentim, who leads data management and AI at IDB Invest, was more blunt about their starting point three years ago. “Our analytical ecosystem was truly outdated. The system made us feel like it was maybe 10 years apart from the private sector.”
The pattern is familiar across government and large organizations: data trapped in Excel files scattered across SharePoints, knowledge buried in PDF reports no one can search, each department maintaining its own databases with its own security policies. When someone needs information that crosses these silos, it requires a formal IT project with procurement, security reviews, and month-long timelines.
Patrick Newbold, CIO of the Centers for Medicare and Medicaid Services, described the stakes in his agency. CMS manages $1.7 trillion in health spending annually and serves 160 million Americans—one in every two people in the country. They defend against billions of cyber attacks each week. Legacy systems aren’t just slow, they’re dangerous.
“We’re not just pairing claims,” Newbold said. “We are a national engine for healthcare.”
The Breakthrough: Business Users Who Build
What changed at IDB Invest happened gradually, then suddenly.
Three years ago, they started building what’s called a “data lakehouse”—a single platform where all their data lives, both structured databases and unstructured documents. They implemented Unity Catalog, Databricks’ governance system, which let them control who could see what data while making it much easier for people to actually access information they were allowed to use.
That was the foundation. The transformation came when they started using AgentBricks.
Kapil’s team needed a cash management system. They manage over 100 bank accounts across 30-plus countries in multiple currencies. The choice was stark: buy an expensive off-the-shelf system that wouldn’t quite fit their needs, or launch a massive custom IT development project.
Instead, Kapil—who describes himself as “a finance guy, not a data scientist”—built it himself using Databricks.
“This was something which we can actually do ourselves,” he explained. “I think the biggest advantage I see—this was first time in my life, I saw the business was doing a lot of stuff and IT was a facilitator.”
Read that again: IT was a facilitator. Not the builder, not the bottleneck, not the reason things take months. The facilitator.
His system uses what Databricks calls a “multi-agent” approach. One agent pulls data from their loan systems, derivative systems, fixed income accounts, and bank connections. Another agent analyzes that data to figure out how to fund accounts and identify excess cash. A third agent provides recommendations on where to invest for the best returns.
The system sends alerts directly to Microsoft Teams when it spots cash opportunities or anomalies. The result: seven-figure cost savings and better portfolio returns. Built by the treasury team with IT providing the platform.
Coming Next
In the next post, I’ll break down the three levels of self-service AI I saw organizations using—from simple document extraction to complex multi-agent systems that combine multiple data sources. Then in the final post, I’ll examine how these organizations maintain security and governance when business users are building AI systems themselves.
