From Policy to Practice: What AI in Government Actually Looks Like Right Now
Our SXSW panel skipped the theory. Here's what a sitting state senator, a state CIO, a top policy expert, and a govtech founder actually said.
On March 18th, USLege Co-Founder and CBO Laura Davis joined Texas Senate Majority Leader Tan Parker, Texas Department of Information Resources Executive Director Tony Sauerhoff, and Texas Public Policy Foundation Associate VP David Dunmoyer on stage at SXSW 2026 for a session called "From Policy to Practice: AI Implementation in Government."
When it ended, the audience rushed the stage and kept talking until SXSW staff had to clear the room.
Here's what they said.
1. The foundation has to come first.
Tony Sauerhoff runs technology for the entire state of Texas. His description of what AI implementation looks like inside government? "A lot less like a sudden revolution, and more like disciplined, methodical modernization."
State agencies run on everything from modern cloud infrastructure to COBOL systems that predate the internet. Procurement cycles run on two-year budget calendars. AI model generations don't. That gap is real, and Tony named it plainly.
The bigger challenge isn't deploying the technology. It's data readiness. In the private sector, you can absorb some data ambiguity if the business outcome improves. In government, you cannot. When AI outputs touch decisions about citizen eligibility for services, the tolerance for error is essentially zero. "Garbage in, garbage out" isn't just a technical problem here. It's a public trust problem.
State agencies run on everything from modern cloud infrastructure to COBOL systems that predate the internet. Procurement cycles run on two-year budget calendars. AI model generations don't. That gap is real, and Tony named it plainly.
The bigger challenge isn't deploying the technology. It's data readiness. In the private sector, you can absorb some data ambiguity if the business outcome improves. In government, you cannot. When AI outputs touch decisions about citizen eligibility for services, the tolerance for error is essentially zero. "Garbage in, garbage out" isn't just a technical problem here. It's a public trust problem.
2. The information gap is bigger than most people realize.
Laura built USLege because she watched this problem firsthand from inside the legislature. The volume is staggering: 5,200 hearings in a single month, 14,000 bills in active discussion simultaneously across all 50 states. For a government affairs team of any size, monitoring all of it without missing what matters is an impossible ask without help.
She gave the room a concrete example. A major data privacy bill kept getting modified due to the number of stakeholders involved. It changed significantly right before the committee hearing. Nobody had time to read the new version, cross-reference what changed, and update their testimony. Nearly every witness testified on the former version of the bill. It passed anyway.
Then she said something that made the non-government people in the room sit up straighter.
"From talking with legislators around the country, not in Texas of course," she said with a playful glance at Senator Parker, "there are states where a legislator has no staff at all. They are truly looking at their colleagues to see how they're voting. That is how bills are being passed in this country. I've been told this directly by legislators. Some don't even have a bill summary."
The room laughed. Senator Parker smiled.
Her point wasn't to embarrass anyone. It was to show what's actually at stake. When a nonprofit advocate has the same access to real-time legislative intelligence as a top-tier Washington lobbying firm, the information gap that has defined this industry for decades starts to close.
"If somebody who's an advocate for a nonprofit has the same access as a top-tier lobby firm, we just gave them a level playing field."
David added a live example from his own work. Mid-meeting with a government staffer, his USLege Radar alert buzzed: the Public Utilities Commission had just announced a report on Water Transparency and Data Centers. The people across the table hadn't heard yet. He had.
"AI did not replace me. It empowered me."
3. Guardrails aren't the enemy of progress. They're what makes it sustainable.
David outlined five tenets of responsible AI governance in government, drawn from Texas's SB 1964 and DIR's Digital Code of Ethics, and argued this framework is a model worth adopting nationally.
Know what AI systems your government is using. Make sure citizens know when they're interacting with one. Apply heightened scrutiny in high-stakes use cases. Keep a human accountable for every AI system. And build real recourse when things go wrong.
He used CPS as the example everyone felt. AI can be genuinely useful for triaging risk across a large caseload and flagging which children need immediate attention. That's a good use. What AI should never do is make the removal decision. "Judge and jury stays with humans."
Senator Parker added his own line in the sand. He'd support AI screening job applicants at the workforce commission level. But when a human being is looking for work to support their family, they need to be engaging with another human being, not just a system.
The throughline from every panelist: AI augments human performance. It doesn't replace human judgment. Senator Parker framed the stakes plainly: "This is the space race of our time. If we don't get this right, America will lose. Future generations will suffer. That's why the framework has to be right."
4. The workforce conversation is more nuanced than the headlines.
Yes, there will be disruption. Senator Parker was direct: knowledge workers will do well, roles with higher redundancy will be more exposed, and anyone who says otherwise isn't being straight with you.
But the more urgent challenge inside state government isn't displacement. It's attrition. The workforce is aging out faster than it's being replaced. The goal isn't to use AI to cut headcount. It's to maintain capacity as the workforce naturally shrinks through retirement and turnover.
Tony's approach to the talent problem: stop trying to compete with private sector salaries for senior AI talent. Build pipelines instead. Bring people in earlier, develop them, accept that some will move on to the private sector, and treat that as a feature rather than a failure. "Better to know and plan for that, and be proud you had a part in helping them go on to do that."
Laura's take was direct. "The future belongs to people who can adapt fastest. It's more human-centered skills that are going to be appreciated. Critical thinking, communication, the ability to keep learning."
David added a call to action for anyone in the private sector with a technical background who wants to make a real impact: the public sector is ready. Agencies know they need the skills. They're ready to trust people who show up with the right mindset.
"If you want to serve your state and your country and see real benefits, now is the time."
The conversation doesn't end with this session. Let's keep it going.