AI Inside the Regulatory Process: Three Practitioners on What’s Actually Working
James Madison Institute Tech Conference Panel | June 4th, 2026
Governments write rules faster than they can read them. The result, over decades, is a regulatory code that grows by accretion, duplicative, inconsistent, and largely unexamined. AI is beginning to change that calculus, and a panel at the James Madison Institute's 2026 Tech and Innovation Summit on June 4 brought together three people who approach the shift from different layers of the same stack to talk not about theory, but about implementation.
The panelists:
Patrick McLaughlin, research fellow at the Hoover Institution and adviser to Vulcan Technologies, who built much of the computational infrastructure, through projects like QuantGov and RegData, that makes large-scale regulatory measurement possible.
James Broughel, senior research fellow at the America First Policy Institute's Office for Fiscal and Regulatory Analysis (OFRA), who spent years documenting the economic costs of regulatory accumulation and now builds the AI tooling that turns that research into operational capability.
Reeve Bull, policy director at the Fulcrum Foundation, who previously led Virginia's Office of Regulatory Management and now works with states including Texas and Oklahoma to integrate these tools into live reform efforts.
The change is speed and cost, not raw capability
McLaughlin opened with a useful corrective: much of what AI is being used for today was technically possible five or ten years ago. What has changed is that everything can now be executed far more quickly and cheaply, and somewhat more intelligently. The underlying problem hasn't moved. If you want to reform regulations in your state or city, you're confronting a pile of text that accumulates year over year and is hard to act on.
What's newly practical is scale. A jurisdiction can now run its entire stack of regulations and produce a comprehensive inventory of how many restrictions exist, how many permit processes, and how many rules tied to guidance documents, in hours rather than the weeks, months, or years it once took. AI also enables more nuanced measurement: not just how much regulation exists, but how many different agencies regulate the same activity, and how that overlapping jurisdiction burdens businesses differently than the same volume of rules from a single source.
The bottleneck moved, from reading to acting
Bull offered Virginia as a test case. The state's reform effort began in 2022, just as ChatGPT was emerging, but deliberately held off on introducing AI until late in the process. The office built its infrastructure and pursued a goal of cutting regulatory requirements by 25 percent, a target it hit in 2025. AI arrived as "icing on the cake," used to tighten wording and surface reporting requirements Virginia carried that peer states did not.
The tools have since advanced considerably. They can now compare a state's regulations against others to flag outsized burdens, run preliminary cost assessments, identify internal contradictions and overlap, and check regulations against the authorizing statute to find where agencies have gone beyond what the legislature actually permitted.
That progress creates the opposite of Virginia's original problem. Where reformers once hunted for things to cut, AI now hands them an exhaustive printout, and an information-overload challenge. As McLaughlin put it, the bottleneck simply relocated: it used to be getting through the stack; now it's acting on what's been extracted. His proposed fix is organizational, not technical. Assign clear ownership, give a specific person the job of identifying what can be acted on quickly, and route that to leadership.
A genuine game-changer for analysis at scale
Broughel, who sits on both the research and tool-building sides, emphasized how much traditional policy analysis can now be done at scale. Ideas once dismissed as pie-in-the-sky are suddenly tractable. One state think tank asked OFRA for a tool to evaluate whether introduced bills violate house and senate rules; a junior researcher coded it and built an interface in about a day, so a user can now drop in a bill and get a detailed report in seconds. On cost-benefit analysis, where the field once produced perhaps ten or twenty reasonably complete studies a year at the federal level, it is now possible to analyze every regulation across roughly 180,000 pages of federal rules.
The real constraint is institutional, not technical
Asked what conditions a state needs before any of this is worth the effort, all three converged on the same answer: people and political will, not better models.
Bull stressed the need for a dedicated body or executive-branch team focused on regulatory reform. Critically, it also needs someone who can bridge the regulatory and technical worlds, because that person identifies the real needs and figures out implementation. Without it, the failure mode is elaborate reports nobody knows what to do with.
Broughel was blunter about the underlying barrier: there is no guarantee anyone uses good analysis. Even the strongest federal economic work routinely gets shelved while politics drives decisions. In his experience, a single strong, motivated, charismatic champion setting clear objectives is often the decisive factor. Bull's Virginia effort and Idaho's reforms are examples. Every president since Nixon has promised to cut red tape; adding "let's use AI" to that sentence doesn't fix anything on its own. An AI-generated report on a shelf is still a report on a shelf. The best version, he argued, pairs a stated goal (a 25 percent reduction, say) with a baseline, clear ownership, and a validation and audit trail, with humans carrying the last mile through the legislature.
What success actually looks like
Two themes emerged. First, a substantially more streamlined and coherent code, reversing the decades-long incentive to only ever add. Bull also pointed to transparency as a major prize: making it genuinely easy for a small business to learn what permits it needs, where to apply, and how long approval takes. That information is largely unavailable today, and it is exactly where AI navigation tools can lower the barrier to entry.
Broughel set concrete markers: fewer pages on the books (190,000 down to 100,000, for instance), an actual accounting of regulatory cost, and reductions of a half or two-thirds, reinforced by institutional guardrails like regulatory budgets, pay-go rules, and sunset reviews. He also named the policymaker's core fear, a public-relations mishap on their watch, and argued reforms must be implemented carefully but pushed through even on a bad news day.
His closing caution doubled as the panel's sharpest point. Economists, he noted, have a habit of asking "so what?" Cutting a code from 185,000 to 100,000 pages is a means, not the win. The real measure is outcomes such as jobs and economic growth, and the public case is made only when reform can be shown to have produced them.
The Monday-morning move
Pressed for one concrete first step, McLaughlin offered a clear sequence: pick a known pain point in your state or city, such as permitting, energy, or infrastructure, and map the full regulatory landscape around that single activity, counting the federal, state, and local rules and guidance documents involved. With that map, propose a narrow pilot, measure whether it improves things, then scale to the next pain point. Start by attaching a solution to a problem people already feel.