We’ve all felt that quiet frustration when we pay our taxes in full and on time, only to suspect that others aren’t doing the same. It’s a nagging sense of unfairness that creeps in every April. Josh Sandler has built an entire company around that very grievance. His startup, Govra, is betting that this feeling can be channeled into something constructive for the public good. Sandler’s central argument is simple: better tax enforcement is the only painless way to fund government services without raising taxes on the law-abiding majority. If governments can recover the money they are already owed—money that has been legally assessed but never collected—they can finance schools, roads, and infrastructure without asking anyone to pay a penny more. For Sandler, it’s not about squeezing taxpayers; it’s about ensuring that everyone plays by the same rules, and that the system doesn’t reward those who skirt their obligations. His mission is to level the playing field, and he believes that the key to doing so lies not in more aggressive human auditors, but in deploying sophisticated AI agents on the government’s side of the desk.
The scale of the problem Govra is tackling is staggering, making the company’s mission feel less abstract and more fundamentally urgent. Sandler estimates the combined federal, state, and local tax gap—the difference between what is owed and what is actually collected—is a jaw-dropping $1 trillion each year. Even the most recent IRS projections, which focus solely on federal taxes, suggest a tax gap of $696 billion for the 2022 tax year, a figure that only drops to $606 billion after accounting for late payments and enforcement actions. To put that in perspective, a single recent audit in Maryland revealed over 18,000 delinquent accounts totaling more than a quarter of a billion dollars that were simply not being pursued. Sandler’s question is pointed and hard to argue with: if a government cannot even collect the revenue it’s already legally entitled to, how can it possibly be expected to deliver the services its citizens rely on? This isn’t just an accounting issue; it’s a fundamental failure of governance that undermines public trust and starves essential programs of the funding they desperately need.
The irony, and the opportunity, is that this collection problem is compounding at a time when government agencies are losing the human resources they need to address it. The federal government has been actively downsizing, with the IRS cutting more than a quarter of its workforce in 2025 under the guise of creating a leaner, more efficient “AI-first” operation. While this might sound modern, the reality is that these cuts come with a huge financial cost. Yale’s Budget Lab projects that the combination of this workforce reduction and a $20 billion funding cut will actually increase the federal deficit by roughly $861 billion over the next decade. Meanwhile, consumers are increasingly armed with their own AI tools that can generate tax appeals and challenge agency decisions with ease. This creates a dangerous imbalance: agencies are losing personnel while simultaneously facing a flood of complex, AI-generated pushback that a depleted workforce is ill-equipped to handle. As Sandler puts it, society is scaling at the speed of AI, but the government is still trying to function at the pace of human workers, creating a chasm that is only set to widen.
Govra’s technology offers a solution that is far from glamorous, but it is deceptively powerful. It isn’t about writing legal briefs or making autonomous decisions; it’s about taking the most tedious, time-consuming parts of an auditor’s job and making them instant. Govra’s software ingests mountains of documents, reconciles financial records, and automatically connects relevant evidence to the applicable law, doing in minutes what often takes human auditors weeks or even months to accomplish. Crucially, Sandler is quick to point out that his software is a tool for human decision-makers, not a replacement for them. It assists in organizing and presenting information so that a human official can make a well-informed, defensible judgment. The system doesn’t choose who gets audited; it simply helps the auditor build a case more efficiently. This grounding in human oversight has helped the company secure partnerships with state tax agencies in California, Indiana, Wisconsin, and New Hampshire. The results are already tangible; Indiana’s revenue department, for instance, estimates that auditors currently spend about a third of their time just assembling the necessary materials for an audit before the substantive work even begins, and they want Govra to compress that timeline from weeks to mere minutes.
The potential impact of this efficiency is significant, but its success will be measured by how it handles the inevitable rise in taxpayer challenges. The straightforward math suggests that getting more audits out the door means more compliance, but it also means the system will face a more adversarial environment. Ashworth, the Indiana official rolling out Govra, is optimistic that they can generate 20 to 30 percent more audits with the same staff. He also notes that the sheer volume of audits could encourage everyone to be more honest on their initial filings. This same technology, however, could also be a boon for taxpayers. It could help individuals contest an incorrect assessment without having to hire expensive professional help. There is great potential for this to democratize the appeal process. Yet, a former IRS commissioner warns of a dark side: poorly constructed AI tools could easily generate invalid claims, clogging the system with nonsense, or expose sensitive taxpayer data. For now, Indiana is bracing for a wave of new, AI-powered challenges to its audits, as a dozen new software companies have already sought certification from the agency, marking a clear shift in the landscape.
But perhaps the most important reframing in this entire story is that better tax administration shouldn’t inherently be viewed as a threat. The tax code is stunningly complex, and a significant portion of the underpayment is not the result of intentional fraud. Ashworth stresses that in his experience, a great deal of what his agency uncovers is the result of honest mistakes made by well-meaning people who simply didn’t fully understand the intricate rules. In this light, Govra’s AI agents become more than just a collection tool; they are a tool for education. By reviewing records for a taxpayer, they can help clear up confusion and prevent future errors. The goal isn’t to punish, but to help people fulfill their obligations correctly. Govra’s success will be judged by its ability to create a more honest and efficient system for both sides of the table. If it can help the government collect what is rightfully owed, while simultaneously making it easier for citizens to comply correctly with the law, then it’s not just a smarter way to tax—it’s a more just and functional way to govern.
