I’ve spent 20 years watching public sector organizations try to get their systems to talk to each other. State agencies, universities, school districts. Different institutes, same problem.
Recently I was on a call with a large education technology organization. Their team is full of integration experts who have spent years maintaining dozens of repositories and the connections that hold them together. Their CTO asked me something somewhat novel and honestly, “ahead of their time”…:
“What if we just put an agent in charge?”
That question tells you where we are. So let’s talk about how we got here.
How we got here.
First, every connection was hand-built. System A needed data from System B, so somebody wrote a script… and then that person eventually retired. Then came integration platforms: better for sure, but with license fees, specialized talent, and multi-year rollouts. Then APIs were next, which mostly finished the job for modern systems.
But the mainframe that is processing claims doesn’t have a REST API. Neither does the student system somebody customized 15-25 years ago. The newest systems connect beautifully. The ones that matter most still don’t.
Where it hurts most: fraud, waste, and abuse.
Nowhere is this more costly than with fraud, waste, and abuse (FWA). The signals are almost always there. Take a provider who keeps billing after being suspended, terminated, or federally excluded… the signal is there, but it’s spread across enrollment, claims, and the exclusion list. Or take a Pell disbursement that doesn’t match a student’s enrollment or bursar account… again, the signal is there, but it’s spread across three separate systems.
Today, a person reconciles those by hand, usually late, and often after an audit finds the gap first. FWA detection isn’t an AI problem first. It’s a reconciliation problem.
The false choice.
Agentic AI has left a lot of CIOs stuck between two options that both feel wrong.
Keep feeding the integration machine.
Keep the team, the licenses, and a custom project for every new connection. It’s multi-year, and your reviewers keep logging into three systems and typing up their best guess.
Turn it all over to an autonomous agent.
Fully autonomous is the holy grail, and I get the appeal. But no state agency is handing Medicaid claims to an unsupervised agent tomorrow, and the people in the room know it. Some of them are also quietly wondering what it means for their jobs.
One option is too slow. The other is too much, too soon.
The in-between.
So we came up with an “in-between” option that is not a platform or a product. Take two, three, maybe four systems. Point an agent at them. Let it join the data, flag the anomalies with a confidence score, and hand back one answer a reviewer can trust. People still make the call.
That’s what we built the Presidio Data Broker to do. It’s an AWS-native, agentic data reconciliation accelerator that lets agencies query across disconnected legacy systems in plain language… without replacing ANY of them.
Your integration experts don’t disappear. They become the people who govern what the agent does. That work is growing, and it’s honestly more interesting work too.
Before you sign with anyone.
Ask three questions:
- Can it work against our actual legacy systems, including the ones with no API?
- Where does a human approve, and where does the agent act on its own?
- How many weeks until our first reconciliation is live?
The Bottom Line
First the answer was more code. Then a bigger platform. Today, it isn’t a fully autonomous agent … NOT yet. It’s an agent pointed at the right two or three systems, with your people still in charge.
We’re going to show Presidio Data Broker live at AWS re:Invent [Booth #520], starting Monday, 11/30/2026.
