Josh Sandler’s pitch begins with a grievance that almost every taxpayer has felt at one point or another: your bill gets bigger when someone else doesn’t pay theirs. It’s a frustration that rarely leads anywhere, but for Sandler, it became the foundation of a startup. His company, Govra, is building AI agents for government agencies, starting with tax audits. At first glance, the idea of more tax audits might sound like a nightmare. But Sandler frames it as a matter of fairness, not punishment. If governments can collect the money that is already legally owed, they don’t need to raise taxes on everyone else. “No one wants to pay more because someone else is skirting the system,” he says. In his view, recovering unpaid taxes is a way to finance public services, pay down debt, or even create room for tax cuts. Govra, which made Forbes’ inaugural Agentic 20 list, is deliberately building for the other side of the counter. While most AI startups are rushing to help consumers fight parking tickets or file their taxes, Govra’s software is designed for the government agencies that receive those submissions. It’s a quiet, unglamorous corner of the AI boom, but Sandler believes it’s one of the most important. His argument is simple: if the government can’t collect what it’s owed, it can’t function. And if it can’t function, everything else falls apart.
Sandler’s path to this idea wasn’t typical. He studied development economics and later built a trucking marketplace in Africa, two experiences that taught him a fundamental lesson: businesses only thrive when governments work. Roads need to be maintained, contracts need to be enforced, and basic services need to be delivered. All of that requires tax revenue. Yet the gap between what people owe and what they actually pay is staggering. Sandler estimates the combined federal, state, and local tax gap at $1 trillion every year. The IRS’s own latest projection, which covers only federal taxes, puts the number at $696 billion owed but not paid on time for 2022. After accounting for expected enforcement collections and late payments, that falls to $606 billion. But even that lower figure is massive, and it doesn’t include state and local governments. A September audit in Maryland, for example, found that roughly 18,300 delinquent accounts totaling $251.7 million were simply not being actively pursued. That kind of money could fund schools, repair roads, or support health clinics. Instead, it sits in the gap between what’s owed and what’s collected. Sandler’s question is pointed and direct: “If you can’t collect money that’s owed, how are you going to provide the services?” He sees this not as a technical problem but as a crisis of governance. And it’s a crisis that’s getting worse, because the agencies responsible for collecting taxes are losing the very people who know how to do it.
The strain on government agencies has become impossible to ignore. In 2025, as Elon Musk’s DOGE swept through the federal government without much resistance, the IRS lost more than a quarter of its workforce. Its leadership has responded by promoting an “AI-first IRS” as a way to boost productivity with fewer people. But the numbers are sobering. Yale’s Budget Lab projects that the combination of staffing reductions and a $20 billion funding cut would increase deficits by about $861 billion over the 2026–2035 period, even after accounting for expected savings. Meanwhile, the rise of consumer AI agents is adding yet another layer of pressure. There are now tools that promise to find deductions, navigate complicated government websites, and generate tax appeals with just a few clicks. These tools make it cheap and easy for taxpayers to challenge their bills. But government agencies don’t have the same luxury. They can’t just fire off boilerplate responses. They have to defend their decisions against legal scrutiny, which means every denial, every audit finding, and every appeal requires careful work. Sandler worries that governments are about to be overwhelmed. “While society’s scaling at an AI scale, governments are scaling at a human scale,” he says. It’s a vivid way of describing an imbalance that’s becoming harder to deny. Taxpayers are armed with increasingly powerful technology, while the agencies on the receiving end are expected to do more with less, and with no room for error.
Govra’s answer to this problem isn’t flashy, and that’s precisely the point. Its software doesn’t make dramatic decisions or replace human judgment. Instead, it does the tedious, time-consuming work that consumes so much of an auditor’s day: organizing documents, reconciling records, and connecting evidence with applicable law. The system is designed to run within the government’s own environment, making it traceable and auditable. Sandler is careful to emphasize that officials retain consequential decisions. The software does not choose who gets audited, and it doesn’t issue final judgments. It simply makes the process faster and more reliable. Govra is already working with state tax agencies in California, Indiana, Wisconsin, and New Hampshire, and it has raised a seed round from General Catalyst. In Indiana, the revenue department is rolling out Govra, and Mark Ashworth, its executive director of business systems support, has a clear sense of what he wants. He estimates that auditors currently spend about a third of their time assembling material before the substantive audit even begins. That preparation can take weeks or months. He wants Govra to reduce it to minutes. The potential impact is enormous. “Can we get 20 to 30% more audits out of the gate?” Ashworth asks. More audits, he notes, might also encourage voluntary compliance. When people know that the system is watching and capable of catching mistakes, they’re more likely to report accurately in the first place. In that sense, the technology doesn’t just collect money. It changes behavior.
California’s Department of Tax and Fee Administration has also partnered with Govra on a pilot program to ingest documents, match records, and suggest audit findings. The department told Forbes it wants to see whether Govra can expand audit coverage without expanding headcount. But there’s a potential upside for taxpayers too. Faster audits mean less interest piling up on taxes that are ultimately found to be due. If you owe money, you want the process to move quickly so the final bill doesn’t balloon. And the same technology could eventually help taxpayers contest incorrect assessments without having to hire professional help. Danny Werfel, a former IRS commissioner who now advises Govra, sees that as a real benefit. But he also warns about the risks. Poorly built tools could manufacture invalid claims, clutter the system, and expose sensitive information. The line between helpful automation and harmful automation is thin, and it depends on how carefully the technology is built and deployed. So far, Indiana hasn’t seen an AI-related increase in tax protests. But Ashworth expects that to change. He anticipates that protests will increase “quite a bit in the coming years” as more people gain access to AI tools that help them challenge their bills. He has already seen a dozen new AI-powered tax-software companies seeking certification with the agency, which is roughly the same number as the ones already on the agency’s roster. That’s a sign of what’s coming. The question isn’t whether AI will transform tax administration. It’s whether governments will be ready for it.
Ashworth believes that better tax administration shouldn’t be treated as a threat by default. He points out that tax codes are incredibly complicated, and Indiana’s audits frequently uncover honest mistakes rather than deliberate evasion. People aren’t trying to cheat. They just don’t understand what they’re supposed to do. “They’re not trying to do it wrong,” he says. “They just didn’t understand what they were supposed to do.” That makes education an important part of the job, and it’s a reminder that tax enforcement isn’t only about catching bad actors. It’s about creating a system that’s clear, fair, and functional. Govra’s approach reflects that mindset. It’s not about building a robot that hunts down tax evaders. It’s about giving human beings the tools they need to do their jobs well, while making the system more efficient and less overwhelming. Sandler’s vision is ultimately about trust. When governments can’t collect what they’re owed, they either cut services or raise taxes, and both erode public confidence. But when enforcement works quietly and fairly, the burden is shared, and the system feels less arbitrary. That’s the world Sandler is trying to build, one audit at a time. It’s not glamorous, and it doesn’t make for a flashy demo. But it might be exactly what governments need to survive the age of AI. The tools that make it easier for citizens to challenge the state are only going to become more powerful. Govra is betting that the state can use those same tools to keep up, not by being harsher, but by being smarter, faster, and more human in the ways that matter most.
