It isn’t every day that the IRS drops something that reads like a blend of legal code, logic theory, and open government idealism. But that is exactly what it did, laying Direct File bare—and releasing it into the public domain—for the world to scrutinize, fork, or admire in stunned silence.
At the heart of this code disclosure is something called the Fact Graph, a dry name for a tool that is quietly revolutionary: a logic engine that parses ambiguous tax scenarios and infers relationships while consistently applying tax rules.
The first line of code for Direct File, written February 8, 2022, was for an experiment to explore the concepts that would become the Fact Graph. Even from that first experiment, the work was motivated by the need to establish trust with taxpayers. Direct File can ask hundreds of questions about your tax situation, but most users will see only a small fraction of that. The Fact Graph provides us with a means of proving that none of the unasked questions would have changed the bottom line of your tax return and that you’re getting every tax benefit to which you’re entitled.
The IRS has been working for years to develop a relatively simpler model of taxes, its “Tax Calculator” effort, which would model the relationships between the lines of the various tax forms. But Direct File and its Fact Graph go further by also tackling the last mile problem of taxes, the often fuzzy relationship between what gets entered on tax forms and the assorted facts about your life that inform what you’re eligible to claim, a relationship that’s variously described via statutes, regulations, court decisions, instructions, publications, and other guidance. This wasn’t trivial, as our team would often be teeing up novel questions for the IRS Office of Chief Counsel to consider as we sought to bring that ambiguity into sharp focus.
This all matters in an age when public trust in institutions is frayed and when algorithms, or artificial intelligence (AI), increasingly mediates our rights and obligations. If the tax code is going to be enforced by machines, and the preparation calculated and handled by the same, the rules those machines follow should be legible and contestable. The open-sourcing of Direct File hints at a future where regulatory logic and automated decision-making isn’t hidden, but published like case law—reviewable, revisable, and maybe even improved by the very public it serves.
Economic Security Project (ESP) announces a new fellowship focused on the future of tax filing, civic tech, and benefits equity. Launching in the wake of the Trump administration’s decision to sunset IRS Direct File—one of the most popular civic tech pilots in recent memory—this new initiative brings together technologists, policy experts, and equity-driven strategists to preserve hard-won progress and accelerate new ideas that make government work better for everyone.
In personal news, I’ve joined with our friends at the Economic Security Project to document and build upon the legacy of Direct File.
The Economic Security Project was one of the earliest and most effective advocates for Direct File, and throughout the program’s existence, they were a leader in getting the word out to the taxpayers who needed it the most. I’m grateful for this opportunity to continue collaborating with Gabriel, Jen, and Merici, reflecting on what we learned and what work remains unfinished.
Speaking of Merici, I’ve been meaning to reupload this video from a July 2024 event, where she and I spoke about lessons learned from the Direct File pilot. The original was taken down in January 2025 due to the transition.
When the Treasury Department ordered the Direct File team to stop work and disband, the toughest part was saying goodbye.
Direct File deserves an Ocean’s Eleven-style montage for assembling the team. The old partner who had to be convinced to come back for one more job (the best, and perpetually most reluctant, user experience lead in the business). The loose cannon I was wary of bringing onto the team due to a checkered past (a brilliant engineer with a crypto startup on his resume). The whiz kid with the precise set of skills needed to pull off the plan (an interaction designer by training, she and I had studied up on tax law together, and she took to it like a fish to water).
We came together, a band of misfits and weirdos, under the banner of a fantastical, improbable mission. It felt like we were unstoppable. And then we were stopped.
And as much as we had accomplished, we knew we were just getting started. So we did our best to document what we had learned. We prepared the project for hibernation, writing notes for a future team that might never come. We said goodbye.
It was August 2022.
It wasn’t Direct File’s first death. By my count it was the second, although there’s an argument for it being the third. Neither count includes all of the doomed attempts to tilt at this particular windmill over the two and a half decades before we took up the lance.
During the initial policy process[1] on Direct File in 2021, I had argued that we would need to make a go/no go decision by June 2022 in order to have the best shot at success in filing season 2024, the final filing season of the Biden administration. And so when the policy process was reconstituted in early 2022, we were told our goal would be to prepare options and recommendations for a June decision. But by May, it was clear that we were going to blow that deadline.
So Merici Vinton, my U.S. Digital Service (USDS) project co-lead, and I hatched a plan. We asked to use our small USDS team to build a “proof of concept.” This prototype[2] of Direct File would illustrate some of the key issues being discussed in the policy process and develop an evidence base to inform future strategy. It would build momentum ahead of a decision, enabling us to hit the ground running if and when that decision came. I said the words “preserving optionality” a lot.
It was also, and everyone knew it, an advocacy tool. As Merici put it, she wanted something so self-evidently good that it would “make it hurt” when decision makers declined to give Direct File the green light. And so a team of three designers and four engineers (if you count me despite splitting time between coding, memos, and meetings), built a proof of concept version of Direct File in eight weeks.
Seizing an opportunity to grow the expertise and capacity of the team was an under-appreciated benefit of this work. Even from this early stage, the product was designed alongside users, and these early insights from taxpayers helped shape our intuitions. The proof of concept served as a laboratory to try out ideas and as a springboard for future iterations. We made mistakes, and learned from them (whatever you do, don’t try to express the logic of the Internal Revenue Code as YAML). The team developed a shared vocabulary, critical for collaboration between disciplines, which the reluctant user experience lead termed our taxonomy. And then there were the inside jokes, and the fridge magnets, and the Pictionary games.
At the end of the eight weeks, we not only had a version of a Direct File, but we had a close-knit group of people who knew how to build a Direct File. Because they had done it.
We began to strategize about what to do next with our small but mighty team. But then the administration’s climate and tax bill, thought dead, was suddenly accelerating toward becoming law. The Inflation Reduction Act (IRA), as it was now titled, would give the IRS an additional $80 billion over ten years. What should have been very good news for Direct File unexpectedly spelled its demise.
The decision was made within Treasury, so I can’t speak to the deliberations that occurred. The stated rationale was that with the imminent passage of the IRA, the IRS was ready to take on Direct File without outside help, and that the effort would be best managed as part of the emerging IRA strategic process. Instead of handing off the work in progress, however, the project would inexplicably start from scratch, walking away from the progress that had been made. Treasury confirmed that they understood this decision would likely foreclose delivery of Direct File by 2024, which, writing from the future where we know the results of the 2024 election, meant that Direct File would never have shipped.
It also didn’t help that the IRA required the IRS to produce a report on Direct File. At this point, no one had a plan for that report (apart from MITRE, which assumed a lucrative contract would soon come their way). The optics of one part of government studying the question of Direct File while another part moved full steam ahead weren’t great. The eventual plan—delaying a decision on Direct File until May 2023 but producing another, more sophisticated prototype as part of the report work and thus preserving optionality (my favorite phrase)—was still a few months off. It turns out that tasking an agency to spend $15 million and nine months writing a report probably isn’t a great way to achieve a result, even if it worked in our case.[3]
Starting over from square one, the IRS was not set up for success. As just one example, one of my top anxieties during this time was how Direct File would handle state taxes. I had written a four-page memo outlining the problem and five potential solutions, the whiz-kid designer had storyboarded out an early version of the experience that would eventually ship, and we had assembled a spreadsheet categorizing the filing options available in every state in the nation. Treasury’s July 2022 memo to the IRS, informing the IRS of its interest in pursuing Direct File, boiled all of this down to the single sentence, “It will be critical to determine the extent to which a federal filing tool can offer taxpayers an integrated way to file their state returns.” Helpful.
More than any memo, though, and even more than any prototype, what the IRS was deprived of was people. Because my top anxiety was: who is going to build Direct File? The IRS didn’t have the capacity to build a product like Direct File, which is not to say that the agency was incapable of it, just that it would take time and leadership to recruit and develop a team with the right combination of skills.
There aren’t very many user experience designers who understand tax law and can effectively collaborate with the IRS Office of Chief Counsel. For that matter, it’s also pretty difficult to find tax lawyers who know how to collaborate with designers. But we absolutely could find talented designers and talented lawyers and give them the space to learn from each other.
These problems can be hand-waved away if the plan is to outsource the work (this was the IRS’s preferred approach). Of course, that doesn’t solve the problem; it just makes it someone else’s problem. It doesn’t take a crystal ball to predict the steaming pile of crap taxpayers would have received after shelling out nine figures (cf. FAFSA).
The work of growing a team ready for any challenge takes time and deliberate care. There are no shortcuts. What hurt us the most was not pausing the building of Direct File but abandoning the capacity to build we were cultivating.
Our team was disbanded. A majority went to the Department of Education and helped launch the ill-fated student loan forgiveness application.[4] The loose-cannon engineer was assigned to a project that made him miserable, but where he would meet a couple of similarly miserable designers with whom he would share the dream of Direct File. Only Merici and I remained, advising Treasury on IRA implementation, hoping we might someday light the beacons and call the dream team back.
Three months later, with interest in Direct File on the rise, Treasury would reverse course and bring our team back to help pull off Direct File from within the IRS. I would be in charge of the team designing and building Direct File, and Merici would serve as deputy to the IRS executive who would run the Direct File program. That’s a story for another time, however. We lit the beacons, and the team returned. And then the team grew.
There were many moments in the course of the following year that I wished we had those three months back. We were comically past the ideal timeline I had laid out back in 2021, and we also now had to write a report to Congress. But regardless of whether this hiatus was death number two or three, it would be the last time Direct File died. Until this year.
I’ve been continuously working in government, one way or another, since 2015. A few weeks ago, I resigned, as did the vast majority of the Direct File team. This time, it was even harder to say goodbye.
I do mourn the fact that Direct File won’t be available to taxpayers next year. But as I told the team as the end closed in, “We took a pipedream, and made it a policy choice.” No one can claim with a straight face that Direct File is impossible anymore; bringing it back requires only that our elected leaders make a different choice.
What I mourn the most, though, is the dissolution of the team, the disregard for the vast impact they were poised and eager to deliver. The team itself is what I am proudest of from my time working on Direct File. Their manic dedication to the mission. The care they consistently took to get it right. The trust and love they had for each other.
One of the team’s many sayings was “the next version of Direct File is the best version of Direct File.” But just as Direct File constantly got better, so too did the team. We were growing our capacity to deliver for taxpayers each and every day, and we were helping to build new capacity across the IRS. Now that progress is interrupted, a reflection of progress interrupted across the federal government, of scientific research interrupted, of international aid interrupted, of all manner of interruptions.
Direct File is a policy choice, but the Direct File team was an orchard, just starting to produce a meaningful yield. Now it’s been razed.
We’ve documented what we learned. We’ve prepared the project for hibernation and written notes for a future team that might never come. We’ve open sourced it for the world to see.
We’ve said goodbye.
Except, well, a member of the team reviewed an early draft of this post, and she would like to refute this point, i.e. my title. According to her, the team has never said goodbye, only “ta ta for now.” We remain as delusional as ever. How else do you think Direct File happened?
It’s a peculiar sort of person who lives through the many deaths of Direct File and says, “Sign me up for one more, please.” But that’s the Direct File team for you. I dare you to try and stop them.
Maybe we’ll be able to relight the beacons someday. And maybe this will be the last time Direct File dies.
Photo: The IRS Direct File team, May 2, 2024.
“Policy process” is a fancy way of saying a bunch of government officials talking about what to do. The process was facilitated via an interagency policy committee (IPC), a working group of people from different parts of government, each empowered to represent their “building’s” perspective. We teased out problems that needed to be solved and other considerations for a decision with the goal of presenting options and recommendations to “principals,” the highest-ranking political appointees. ↩︎
Not to be confused with a separate prototype that was built in 2023 under the aegis of the Inflation Reduction Act-mandated Report to Congress. The later prototype was significantly more robust, and it built out the codebase that would eventually ship to taxpayers. ↩︎
I like to say that the Direct File pilot cost just tworeports to Congress, and we threw in a tax filing tool for free. Oddly enough, $15 million being a reasonable price tag for a report is something on which both parties agree. ↩︎
In the Who Cares Era, the most radical thing you can do is care.
In a moment where machines churn out mediocrity, make something yourself. Make it imperfect. Make it rough. Just make it.
At a time where the government’s uncaring boot is pressing down on all of our necks, the best way to fight back is to care. Care loudly. Tell others. Get going.
At last week’s Code for America Summit, Suzanne Chapman and Katie Aloisi spoke about how Direct File was designed to build trust. Katie spent a significant portion of her time getting into the weeds of the tax situation of a fictional multi-generational household.
The audience wasn’t bored. They were riveted. Suzanne and Katie’s care was self-evident and contagious.
This was in contrast to an increasingly platitudinous atmosphere. The same recycled stories, making the same high-level points. And I generally agree with the points! But where the details haven’t been completely glossed over, sometimes the speaker got them wrong.
I’m by no means innocent on this count. During a panel, I recited a story I’ve told many times before. But Dan’s piece has been rattling around in my head as I think about how to approach writing in public for the first time in years.
Caring requires more than saying the right things. It requires more than telling someone, “I care.” Care is shown.
…the team continued to evaluate Direct File, as it had all along, preparing a comprehensive year-two report on its work. We filed a Freedom of Information Act request for that evaluation, and this week received Direct File’s 2025 season-end filing report. It tells an inspiring story of a product genuinely transforming tax administration in America, a story made bittersweet only by the fact that such progress has been cut short.
In addition to their summary, the Center for Taxpayer Rights has posted the full report for all to read.
