When FAFSA Broke, They Called This Guy

Jeremy Singer is the President of College Board, which he has led for over a decade. In that role, he oversees the SAT, AP, and other core elements of the U.S. college access ecosystem, and he’s previously had leadership roles at Kaplan and McGraw Hill Education. Why is Jeremy on Statecraft today? After the failed redesign of FAFSA in 2023, he spent six months at the Department of Education helping to ensure the 2024 launch was successful. The revised application form meant 1.7 million more students were eligible for maximum Pell Grants in the 2025-26 application cycle. We discuss: Why attempts to simplify FAFSA went so badly wrong The problems caused by precise drafting in Congress How Singer got FAFSA back on track What politicians and GAO don’t understand about developing software Thanks to Harry Fletcher-Wood, Shadrach Strehle, and Jasper Placio for their support in producing this episode. For a printable PDF of this interview, click here: Our story of the day is the salvage operation you did a couple years ago on FAFSA, the Free Application for Federal Student Aid. But first, my most urgent question. I’m sure that in your role as president of College Board, you saw my SAT scores. What did you make of them — room for improvement? Excellent scores — well-equipped for the modern world. No, that is a very common question, but we do not have access to scores. You don’t dig around? Even very important people, we cannot access. You’ve been president of College Board for a while. It’ll be 13 years — the longest I’ve ever been in any one place. That period was punctuated by a six-month stint in the federal government, and that’s what I want to talk to you about today. Around 2020, there was a big bipartisan bill to simplify FAFSA, the main system for financial aid for college applicants — to go from about 100 questions to 36. When the new system was finally rolled out, at the very end of 2023, it was totally botched. Tons of applicants couldn’t access the site or fill out their forms. Colleges had to wait months to get initial financial aid information from students. For a number of your readers, it will bring back PTSD. My six months — I call it the world’s worst sabbatical. It was good to have a bit apart from College Board, but it was jumping into a fire. Let me start at an even higher level. There’s a ton of research — Raj Chetty out of Harvard is one of the best — that shows a college degree is the surest pathway to a better life. For a huge number of students, FAFSA is the tool that unlocks financial aid allowing them to fulfill that — go to college, get a degree, and improve their pathway. How many American students fill out FAFSA in a year? There are roughly 40 million people accessing the system on an annual basis. But that’s both students and families. I think it’s at least 17 million. If you’re familiar with federal Pell Grants — FAFSA unlocks Pell. There are federal loans you can get; there is Work-Study, where the feds pay you to work in college. But FAFSA is used by many states, and some colleges, to open up their aid to students. It is the linchpin of financial aid. [Editor’s note: the transcript originally said 17 million people access the system on an annual basis — it’s actually 17 million students, and 40 million people (both students and their families).] So I fill out FAFSA as an applicant, that’s a federal program, but then a state like Louisiana needs that information to prime its own financial aid system. Exactly. Students could either be an independent student — they’re no longer dependent on their family for financial aid — or they’re dependent, like many younger students are. A prospective college or higher-ed student and their family will complete the FAFSA form. It will be analyzed to determine what level of need they have, based on their income, family income, and all these other factors. FAFSA generates something called an Institutional Student Information Record, generally known as an ISIR [pronounced “icer”]. The federal government sends it to states and colleges. Based on that, they will know what kind of federal aid the student is going to be eligible for. But in addition, if Louisiana is giving state aid to students, they could use the ISIR to determine how much to give that student if they go to Louisiana State. And many institutions that give aid to students in need use FAFSA as the best proxy for that need. It’s all contingent on the information originally filled out by the student and their family. Which millions of kids are filling out every year. Before FAFSA was redesigned in the last couple years, there was good research that there were literally millions of students who would be eligible for federal and potentially state and institutional aid, but never completed the FAFSA. It was too complex, they didn’t know about it — there’s a lot of reasons. This had been an issue for a long time. [Former senator] Lamar Alexander deserves a lot of credit. He had a lot of desire to simplify this process so it’d be easier and more accessible — many more families would be able to complete it, and it would open up college to more low-income students. When he was on his way to retire, in 2019, there were two congressional bills about FAFSA simplification. The question was, “Can we take this cumbersome, 100-plus question form process” — a lot of it was taking data from your tax returns and figuring out field 38 needs to go here, and a family inputting it and potentially making mistakes — “and make it easier?” There would be logic built into it. Think of TurboTax — when they find some information on you, they say, “These six questions are no longer relevant, so you no longer need to ask them.” The single biggest breakthrough was this was going to pull data from the Internal Revenue Service (IRS). Santi, you did a tax return hopefully? You’re making me second guess. Did I leave the oven on? I’m trusting you here. Let’s say you filed. You have a child. Now, instead of you trying to figure out what parts of your tax return you need to physically enter into the form, it would ingest that from the IRS. That dramatically reduces the time and amount of input — families literally go from hours to complete the form and find all the information, to tens of minutes — some under 10 minutes, depending on the complexity of the individual. The instinct behind that bipartisan congressional push was, “The federal government has my tax information. There’s no reason we should make citizens go dig that information up again. It’s just a matter of linking up these massive internal federal systems.” That’s right. All well-intentioned, all made sense. But it became the biggest software project that the Department of Education (DOE) had ever done, and it was not initially successful. It was supposed to launch in October of 2023. It launched much later, and it launched in pieces. You were brought in in June 2024 — six months after the botched rollout. You were responsible for running the cleanup job. In 2019 and 2020, Congress told the Department of Education, “You’ve got the better part of three years to set up this new system. We want it ready for kids in the fall of 2023 to apply, so that in the fall of 2024, they go to college with however much in federal aid.” What happened between those bipartisan bills passing to great fanfare, Lamar Alexander retiring, and this nightmare rollout? I’ve spent a lot of time trying to uncover that. I got a call from the White House in May 2024. They were modeling it after the Affordable Care Act. They launched the exchange, they weren’t working, they kept trying to fix it — eventually President Obama had the foresight to bring in a team from the private sector with software experience. They fixed it. This was the same. They needed people with large-scale software experience, and my name got floated. The students that couldn’t complete FAFSA are the students we were trying to serve at College Board. Also, we’re a membership organization, every higher ed institution is a member, and it made their job next to impossible that year. We decided the best thing I could do — more than anything I could do at College Board — was to try to right FAFSA. I called every brilliant person I worked with in my career: friends, colleagues, former colleagues. I had a great team of eight people. Jeff Olson, who’s the Chief Technology Officer (CTO) for College Board, came; he said, “You’re way over your head. You need me.” But other brilliant people — including Aaron Lemon-Strauss — came on board. When I left, he stayed, and he’s now the GM of FAFSA, which is essentially what I was, with Jeff being the CTO, which they’d never had. They’d never had a CTO? There’s the Department of Education, then within that, there’s Federal Student Aid (FSA). There are a couple of organizations set up by the government to be performance-based. They want it to run more efficiently — the idea is to run it more like a private sector. Why is that? There’s trillions of dollars of loans — it’s one of the world’s largest lenders. It’s in the DOE, but it’s really a huge financial loan organization. Within the FSA, there was a Chief Operating Officer (COO), which in government they pronounce “Coo” — I’d never heard it referred to as that before. The COO was supposed to be the person to run all of FSA. FAFSA’s a big piece of it, but there’s a lot of other pieces, like how you manage outstanding loans. The COO should have been the GM of FAFSA, but they had a lot of other responsibilities — there was no one running FAFSA. There was a CTO of FSA, then there’s someone at the Department of Education, but not specific to FAFSA. That becomes a big issue. What went wrong? I’ll bucket into three large areas. One, the two bipartisan bills that passed in 2019 were incredibly well-intentioned, but were overly prescriptive on how the system had to work. They hard-coded product requirements. 40 years ago, everything was what‘s called “waterfall”: There’s a team that defines exactly what the output has to look like, They hand it over to developers to code it, They hand it over to quality assurance who check it. That process, while very large in the ‘80s, was replaced, in most cases, by some version of agile. Agile — the idea is, your future self is almost always going to be smarter than your current self. If you try to define everything upfront, you’re going to learn in the process and you’re going to miss out if everything’s hard-coded. Instead of trying to define everything, you start iteratively producing code, getting feedback, learning, and that loop keeps going. For some projects, where there’s a very specific deliverable, on a date, to a budget, it’s hard to do agile. Agile — it’s harder to predict those dates. People will use waterfall occasionally. If you’re going to build a skyscraper — we both live in New York, there’s the big skyscraper in Manhattan that is waving. That was a waterfall. They tried to define everything. You have to. You can’t say… “We’ll figure it out 50 floors up.” Exactly. Where the ducts are — that all has to be defined. What you do with the foundations depends on what’s going to be constructed. Buildings are waterfall, but many other things don’t need to be. It’s been interesting talking to my colleagues at IFP on the infrastructure team about what better planning looks like. In some domains, what you’d like to see to build products more efficiently is more upfront planning. But that’s in places where you get one chance to dig. The budget bloat in things like the Boston Big Dig is a result of not understanding that you won’t get to go back and iterate, because of the physical constraints of the process. Whereas in software you can write new code — you are allowed to learn from your mistakes. 100%. What Governor Shapiro did in Pennsylvania, remember the I-95 and how fast they fixed that. We’re stretching it, because it’s not software — but it wasn’t agile. “How do we get it fixed correctly, quickly?” Maybe there is some application in certain instances. It’s more complicated than my simple gloss. But even with high-speed rail in California, there was not a lot of planning inside the California Department of Transportation. They didn’t have the technical capacity internally [as Zach Liscow discussed on Statecraft]. As a result, as you realize things you didn’t know about the project, you have to go back after you’ve started. I’ll give a very specific example. There was a very well-intentioned Democratic senator who cares a lot about the unhoused population, particularly unhoused applicants for FAFSA — “unhoused” being the term of art for “homeless.” In the bill, they said, “You have to give unhoused people more scaffolding to complete FAFSA, because they don’t have a home address to list.” 100% makes sense. But the way it got defined, the team that was executing had to put this interstitial direction that was specific to a percent of a percent of FAFSA applicants were unhoused. I’m not saying they’re not important, but everybody had to see that language. If you weren’t unhoused, it was — “What are they trying to say? Do I not list my address?” It was well-intentioned, but we quickly saw so many students got confused by these pages. But because of the way the statute was written, FAFSA couldn’t take it out. Similarly, we had a Republican senator who was very active in writing the bills. I was with her in front of a student filling FAFSA out when we were doing beta testing. She was… not a fan of us. When we’d first come, she’d told the press, “There are people from College Board that are going to further torpedo the FAFSA to help College Board somehow” — that we were there to be saboteurs. But in the beta testing, there was some confusion in the software and she turned to me and — credit to her — she said, “Oh my God, this is something I put in, not knowing that this would be confusing to the user.” You could have got that information in a much less confusing way. She immediately saw it because she saw the student struggling with it and she’s — “I didn’t think it could be interpreted like that.” So that’s [problem] one. The legislators — their view is, “We can’t leave it up to career staffers to define all these terms.” But then they’ve handcuffed them. The second big issue is, who’s producing the software? This was the biggest software project in the history of the Department of Education. They had limited deep technical experience in the department or in FSA. When you say limited technical expertise, put some numbers on that for me. How many engineers were there at FSA? There’s probably a decent number. But they had four primary software vendors. General Dynamics Information Technology (GDIT) was the biggest. They had a very large contract with Accenture, and then two smaller tech firms. My father worked at IBM back in the day. The idea — “You never get fired for hiring IBM” — probably in government you rarely get fired for hiring Accenture or GDIT, because they’re big monoliths that the government uses a lot. “Beltway bandits” is the critical term people use to describe these big consultant firms. I don’t know how good these firms are technically, but — GDIT had built the old FAFSA software. It was built in archaic software languages that were no longer being used — COBOL — and on mainframes. A lot of the people we were getting were the same people that had built the system and didn’t know Python, or modern software techniques. They were ill-equipped for the job. What was clearly needed, and I think what Aaron Lemon-Strauss has successfully built, is 15-20 senior technical people who understand the architecture, and can effectively keep the vendors accountable. There was no ability to check the veracity of what the vendor said as far as status, quality of code — all the things you’d want to do. They had four large vendors. The vendors did not work well together. There’s no perfect solution. Sometimes government [chooses] one big contractor, but then you’re hostage to that contractor. They know everything, they control, and you really can’t do much. The argument for giving it to one contractor is, “You have one neck to squeeze when things go wrong.” You say, to Accenture or whoever, “we’re holding you to account.” That is the argument. Otherwise you get five people pointing in twelve different directions of who to blame. A ton of people were our guides. We all had software and education experience, but none of us had government experience. Jen Pahlka [who Statecraft interviewed in 2024] was our guru; her book was our Bible. Danny Werfel from the IRS was willing to meet with us and advise. He was fantastic. His take on this is, “It’s costly, but if you’re going to pick a single vendor, keep a second company on the side, and make it known.” Pay to keep the second vendor up to speed so that you have a credible threat to the first vendor: “If you don’t do this better, I’m going to switch vendors.” There’s a cost to that: you’re constantly paying a backup to be ready to come in. You have a backup quarterback who’s always doing reps. But then you can say, “If you don’t do better, I can fire you.” If I was going to do one, I would do that. The other model is where you have four or five vendors and they’re all involved. If one’s not performing, you move it over. Procurement has all of its own issues: there’s archaic rules, it becomes political, it’s a mess. But the other issue is, if you’re going to have four or five vendors, you need systems that allow them to work together. When you release software, it’s an evolved process to make sure it’s going to perform the way you want. When they wanted to release code, each of the four vendors had to do something, but they had to do it independently. They weren’t on the same systems. They were emailing or texting things that should be in a system. One of the vendors builds the security verification, one of them builds the plugin to get the IRS data, and one of them builds the front end that I see as a 17-year-old applicant? [Another does] the analysis to determine what the score is so they can generate an ISIR, another to generate the ISIRs. It’s all those things. None of this is visible to the user, but it’s essential for it to work well. You try to get a common interface so that the ability of systems to talk to each other is easy. Everyone talks about that — easier said than done, but you use APIs, so you know what to expect if you’re one end of software ingesting information from another. Equally importantly is something as simple as communication. That was terrible. After I left, we got a very happy note from Aaron Lemon-Strauss to the group that had worked on this, that they had finally all gone on Slack, which was hard. We were trying to get everyone on the same system. All the vendors? All the vendors and the FSA. That was a huge breakthrough. It sounds so freaking obvious. If you’re in the private sector, obviously everybody working on a software project would be on the same system so you could communicate. Imagine — “I got this in Slack, I’ve got to copy it to Teams so this person can hear.” It’s absurd. The third [issue] was organizational within the FSA. Some of the classic — unclear decision rights, escalation took a long time, very risk-averse culture. On both parties a lot of politics gets into it. All this compounds the problem. What was interesting — there was a train wreck happening. FSA, Department of Education, the White House — I don’t think anyone there realized. Remember, October 1st, 2023 is when they’re supposed to launch. My guess is spring of 2023, the vendors were — “We’re doing well.” It was completely non-transparent to the department how screwed they were. If I was doing a release on October 1st of that scale at College Board — or when I used to be at Kaplan or McGraw Hill — we would’ve been code complete probably in May 2023. Then we would’ve been just testing — load testing, user testing, improving on the edges. October 1st when you open to 17 million people, it works. They were still building core functionality very late — even past October it turns out. I don’t think the department knew how screwed they were. They were somewhat surprised as it got closer how bad it was. Even then, they didn’t know how bad it was. The well-intentioned people in the department and the White House — if you don’t have the information, you’re reactive. They pissed off the community, higher ed, the Community-Based Organizations involved, the high schools, the general public. Because they couldn’t adequately communicate the situation, and they kept putting stuff out that ended up being wrong. I don’t want to make this a blame exercise — I’m more interested in your perspective on the cleanup. But for my own personal education, I’m curious to understand how something like this happens. The healthcare.gov debacle and then that successful cleanup happened two administrations prior, in the Obama years. I’ve always understood that as the beginning of a sea change in American policy around how you build tools and service delivery. What meant that those painful political lessons weren’t learned by the next Democratic administration? I don’t think they were absorbed by either party fully. Developing good software — particularly that has the complexity of FAFSA or other software that the government does — is not easy, even in the private sector. There’s a lot of private, well-run companies that botch software launching. I’ve had a few in my career — not to excuse it, but it’s a hard thing. If you’re going to build software, what do you do? The alternative is don’t use vendors, build it yourself. Some have tried that in government. That’s really hard. Can you attract the best software talent? You’re competing with Google, Facebook, and private enterprise. There are brilliant people who could do better in the private sector, but want to have impact. But, by and large, you’re going to get a lower quality of person who’s going to see this more as going through the motions. So it’s very hard to build it internally and support it long term. The hope is you can find that smaller slice of people with very strong technical [knowledge], which they didn’t have. Part of what came out of the healthcare.gov story was this idea of trying to use multiple vendors. But it’s not that simple. They’d say, “Use multiple vendors, but you need clear, expert leadership.” If you don’t have the technical expertise and you want a single breakable neck, there is a world where you say, “Accenture, you’re going to be over all these vendors, and it’s your job to make sure it delivers.” Then you could have gone to one org and said “You screwed it up. You’re accountable.” But they didn’t have either of those. It’s my understanding that there were no product professionals at FAFSA in the fall of 2023. But you’re slowly shaking your head. There were people that were product experts in the old FAFSA system. They understood workarounds and the minutiae — because there were some things that the old software had issues with and they knew that stuff inside out. Are those the people I would put on rethinking the software and developing something modern? No. Like the GDIT engineers that were COBOL engineers and knew how to code something in 1980 — these people were not the people I’d want to develop an innovative product in 2023. No question they didn’t have the right people on staff, either in FSA, or the department. Then it became a crisis. A lot of well-intentioned people — smarter than us in policy and many other things — jumped in to try to fix it. But they also were not software or operational people. They didn’t know what they didn’t know. They made some improvements, but they couldn’t figure out how to hold the vendors accountable. A big part of my job was trying to keep some of these well-intentioned people, who didn’t understand software, away from the software — to give the rest of the team the opportunity to do work. We were lucky. There were United States Digital Service people — this is like a civic Peace Corps. They are very successful technically — engineers, product designers, data scientists. They come to government from the private sector and do two or three-year stints — helping veterans get benefits [as Marina Nitze has described for Statecraft], you name it. Very effective. Jen Pahlka, again, one of the origins of that. There were about 200 of them in the government before this current administration. We had about 15 working on FAFSA. But they were frustrated because they didn’t have room to do their work. You had politicals, the White House, suddenly hyper-engaged on this front-page news story. How did you give your team cover? It was a lot of selling. I was lucky because it was an Intergovernmental Personnel Act (IPA) agreement. College Board was still paying me and let me work for the government. I was of the mindset — “You brought me in to fix this, and if you don’t like it, fire me. I’m happy to go back to my College Board job. I’m not enjoying this work.” It was a brutal amount of hours and stress. You had your own other vendor lurking in the background — “I can walk at any time.” I was a bit of an asshole. But I brought a lot of credibility with higher ed, with K-12, with Community-Based Organizations (CBOs), that are essential to FAFSA. Will you define CBOs for me? Community-Based Organizations — these are nonprofits that are part of the ecosystem that helps families and students complete the FAFSA. They ride shotgun, they go to schools and communities to help them complete it — and are so essential to how it works. The National College Attainment Network is the probably biggest. Part of the success we had was thanks to all those organizations. When I got there, there was such a fractured relationship. To your question about what I was doing — we ended up meeting weekly, a call with staffers for the top four Democrats and top four Republicans who work on education in the House and Senate. It was a group of 20. They’re very frustrated, because they wrote the bills and now the software wasn’t working. I was the face of that. They were unpleasant calls. Similarly, we were in a weekly White House call, and we met with the Secretary. Here’s a great example. We got there in June, and realized there was no way we could launch the next year’s form on October 1st. If I was at any of my old software jobs that significant, I’d probably be code complete in May to launch something to this size population. They were not only not code complete, they were still fixing bugs from the prior year. There’s no way we can hit October. But the system was all going toward an October 1st launch. In fact, while the department knew that was a Hail Mary, there was pressure not to acknowledge this. Within the department? Within the department, the White House, the community. We had these well-intentioned associations of college presidents, high school associations, and community-based organizations — they wrote a letter to the Secretary and Congress to say, “You have to launch on October 1st.” I get there and — there’s a phrase in software: “Two moms can’t produce a baby in four and a half months.” There are times where you can throw the kitchen sink — as much money, whatever, and you still can’t produce quality software in a certain time. I said, “This is a suicide mission.” The last thing I wanted to do is work my ass off and then all these constituents that College Board serves, that I know — I’ve built credibility over my whole career — disaster. All the constituents wanted October 1st, but October 1st is not a magic date. It used to be January 1st, and they moved the deadline forward maybe 10 years ago. Earlier is better, but it’s not like November 18th, when we launched, was a disaster. So many colleges, high schools, and CBOs do stuff to help students complete the form. They told me, “If it’s not October 1st, we’d much rather know a firm date that you’re going to hit, and it’s going to be quality software, so we can plan around that. If there’s a 70% chance you’re not going to open October 1st, but you don’t tell us first, that’s a disaster.” I couldn’t convince the White House or the Secretary. The Secretary had gone in front of Congress in May, and apparently he’d been told, “Do not commit to October 1st, no matter what they ask you.” They knew there was a real risk. Of course, in the moment, he says, “We’re going to make October 1st.” This is Secretary Cardona? Yeah, Miguel Cardona. The bright lights of the oversight hearing are just too hot… It’s hard not to. You want to do good. I get in there, I’m trying to convince the White House, and they’re worried they’re going to get crushed, because they have this letter from all these constituents. I start calling all the people I know and say, “We’re not going to be able to do this. We can pretend, try, and miss, but it’s going to be very similar to last year.” The first year — they claim it was December. It was a bit December, but it was into January. But that was only the part of the software where the student and family could enter their information. The production of these ISIRs, which go to colleges — that functionality didn’t open until many months after January 2024. When they produced the ISIRs, they were “Oh s***. There’s an error in the submission process that is generating faulty ISIRs.” Millions of students had to go back in and redo this — now we’re talking spring ‘24. We knew that if we tried to go for October 1st, at best, we could only do that front end again, where people could submit — and there was a chance they’d have to redo what they submitted. That would be a disaster. But it took so much gamesmanship, political capital, and triangulating to get everybody to accept that. We eventually did. My role was — not just in the government, but with the community, the whole ecosystem — trying to help them be realistic. Because all these people were aligned. Everybody wants FAFSA to work. Republicans, Democrats, higher ed, CBOs, high schools. I’ll give you another example. There was a situation where we ended up doing beta. Whenever you’re doing software, before you open something to 17 million people, you’d want to test it with as many people as possible, so that you can find and fix bugs — so that it works. There’s a rule of thumb that if there’s an issue, if one user in every ten hits that bug, you usually will be able to diagnose and fix it if you have 100 users. You want 10 people to experience that bug, then you can probably figure out what it is. A 1 in 10 bug is a disaster. But we started our first beta with a few hundred students, so that we could say, “If there’s that frequent a bug” — luckily there wasn’t, “we would find and fix it.” We found little things that we did fix — usability issues. Then the second beta, two weeks later, was thousands of students. Now we’re finding 1 in 100 issues. Our last beta was tens of thousands of students. We would find 1 in a 1000. We weren’t going to find 1 in 1 million — when you have 17 million people, there’s still going to be bugs. That six weeks of beta testing was essential. We went out to schools, we watched them complete the form, and we discovered so much. I took one of the FSA product people that had been there 20 years. It was her first time she was watching a user interface with software, which was very upsetting. It’s a week before we release. Whenever you do anything to the code, there’s a risk that you create some other issue. There’s something called regression testing, which is supposed to help reduce that risk, because you test the software in a million ways, but it’s not foolproof. I err conservatively. If I have a system that’s working, I don’t want to mess around with the code a week before it’s going to launch. We introduce too much risk. There was a small population, less than a percent of users — they could complete the form, but it was not a great user experience. It may have been incarcerated students. We were like, “We’re going to defer [fixing it for them] — because we’re about to launch 17 million people.” The last thing we want to do is accidentally introduce something that affects all 17 million. I got a call from someone very senior in the government who said, “You have to fix it.” I said, “No, it’s too risky. No one in their right mind who knows software would ever play this game. We’ll fix it later.” We went back and forth, and finally I said, “My CTO and I say this is a disaster. But if you want us to do it, all you have to do is send me an email saying that despite our recommendation, you know more about software, you’re telling us that we should do this, and if something goes wrong because of this…” Of course finally they back down. But there’s a lot more of that s*** I had to do than I wish I had. If I ask you for the name of that very senior political official, I don’t think you’ll tell me. But what’s the larger lesson there? Politicals don’t want to put things in writing? Both parties face such pressure now that it’s not very conducive. One [lesson] is, you’re bringing your experts for a reason, and you’ve got to listen to them. This person did incredible things, but in this one area, they were out of their depth. The thing that’s depressing is just how political all this is. Let me give another anecdote. We had to pitch House and Senate staffers on why beta makes sense. There was a lot of pushback at first: “It doesn’t make sense, you should just launch.” Finally we laid it out in such a convincing way. The staffer I mentioned earlier, for a prominent Republican senator who’s still there — that staffer loved beta when she finally understood. She said, “Can you help us figure out a way to put a requirement in any government software that they do a beta launch before they launch fully.” This is the same person who’d told us we should quit, because we’re from College Board and we’re trying to undermine the government. You won her over. We were so happy. I’d rented an apartment in DC for the six months. We went to the apartment, and we were having dinner before we did more work. Her boss, the senator, tweets out, “Failure of the Biden administration. This beta thing is just an attempt to confuse the public.” Totally had weaponized this in a bad way, despite his main staffer loving it so much that she wanted help. It’s stuff like that — your head just explodes. When it came to student loans, the Biden admin had a big other political priority, which was loan forgiveness. Over the course of 2023, and especially ‘24 when this rollout got botched, there were a lot of allegations from Republicans that the Department of Education politicals had taken their eye off the ball, and that focusing too much on loan forgiveness led them to be poor stewards of FAFSA. How much credence do you give to that claim? It goes deeper, to what you think the role of government is. If your maxim is, you want government to help citizens of the United States succeed, including experience quality education, get a degree, end with an affordable amount of debt that they can repay, and have a system that makes sense in how they repay it — Democrats are more ambitious on the things they think government can do. I’m not breaking any news here. If you’re in the Republican party, and you’re specifically very minimalistic on what government should do, you have a lot less ambition. When you talk to the lifers at the Department of Education, they are more inspired, on average, by what the Democrats want to do. But they sometimes have more success in Republican administrations, because Republicans have many fewer priorities. To that extent, there’s some veracity that the Biden administration tried to do too many things at once in education. They were living through COVID, so we could debate, “Were most of those things required?” My intuition is that it’s very easy to take your eye off the core functions. This was a bipartisan, statutorily-required thing that FAFSA had to do. It was an incredibly big, ambitious modernization — but it’s also something that needed to get done. I don’t think people understood the complexity and magnitude of it — which happens all the time. It’s easy to look back and say, “It was such a big project, it should have been the priority.” But it’s also hard to know — they were dealing with a million curveballs that COVID created. Easy for us to sit back and say, “I would’ve just focused.” In the moment, it would’ve been a very hard thing. I want to change gears and hear your perspective on the Government Accountability Office (GAO), which issued a very critical report in 2025. The report criticized the botched rollout, as well as your cleanup operation. Your colleague Aaron Lemon-Strauss, who is still there, wrote a response to GAO saying, basically, “No, you’ve got it wrong. You don’t know anything about building software. If you did, you wouldn’t say this stuff.” [Good piece from Jen Pahlka on the GAO response here.] Everyone should read that response, because it’s genius, and it’s 100% right. I was never involved in government, so this was wide-eyed Jeremy going to DC thinking about how things work, then seeing the reality, and being — “Oh s***, this is how the sausage is made.” The biggest hit was GAO. I have always viewed them as smart — the place that keeps everybody accountable. Who doesn’t like accountability? Exactly. Then I see it. I’m like “Holy s***.” If we had followed what they suggested, it would’ve been a much larger disaster. It was almost as if they programmed Claude, or whatever they used, in 1990 practices — “This is how you make software” — to give a recommendation. Because it was such a mess. Oversight’s a critical function, but the GAO team that did this was so bad. It was very compliance-oriented. Compliance does not lead to successful software. What did they ding you for on the process and compliance stuff? There was stuff like, “If people had taken certain trainings, we would’ve avoided this.” They were critical that in these moments of crises — prior to me — people didn’t go through the right steps to document everything. But if you’re dealing with a house on fire, you’re not going to go, “I’m going to first go to the second floor” — it’s detached from any reality. They were requiring key people that I needed to launch the software in the fall of ‘24, and they were having us produce old documents and review stuff and I’m like — “Guys, we need to do this.” To the credit of the department, they tried to help make sure that the critical staff we needed didn’t get too distracted by these exercises. But when I talked to them, it seemed they didn’t get it. It’s not people I would want to give a recommendation on how to produce good software. These are not unique criticisms of GAO. I’ve heard similar complaints from people in different domains (notably the CHIPS Program Office). I think their framework is broken. I also wonder whether they have the quality of mind, of people, that can understand there’s a framework, but also understand nuance and say, “You do want good documentation. But when you’re in an instance like this, that’s not possible. I’m not going to ding them for that. But what I would’ve wanted to see is X.” Or understand waterfall versus agile and say, “I get it, there’s less documentation because of this. But if you’re doing agile, you should be doing these six things that are essential for agile success.” You guys had Congress breathing down your necks after the failed rollout. What would a good version of that look like? One that did all the things you want out of oversight, stopped you guys from wasting money or fraud, and kept you on mission? I don’t think there’s a lot of fraud involved. No, but that’s a big reason we have this whole oversight system. I don’t know how we do it, but if we could depoliticize things, that would help. It’s not normal that, on a project like this, the team reports to 20 congressional staffers once a week. That was a reaction to the early issues. It wasn’t a bad thing for Congress to want to be closer to something that had issues. But I wish it was less political. There were moments where they could appreciate the good thing and ask questions. It didn’t have to be as staged or theatrical as it was. Jeff and I benefited because we didn’t want careers in politics. In fact the opposite. So we could be honest. We didn’t care about our brand to some of these people. If we did, we would have to be even more political in what we said and how, and maybe we don’t share as much. We had this whole concept of working in public. Aaron Lemon-Strauss started this — we wanted to put up a blog — because the ecosystem’s so large and we wanted to galvanize them. We said, “We’re going to work in public, we’ll talk about what bugs there are, so people can help their students, or work around them.” We got such resistance to that openness, because there’s a danger: the more information you give out, the more stuff that can be weaponized politically. I wish that wasn’t there. There would be a tendency for more openness. What could Congress have done, in writing the statute, to avoid some of these failures? In software, you develop use cases, but you don’t specify how it works. A use case could have been, “I want an unhoused applicant to easily understand what they should do in this section that is complex if you don’t have a house.” You could’ve defined it like that, as opposed to, “You have to give this language to people.” There are a million examples where you could say what you want the software to do, then let the team figure out how to do it most effectively — test and iterate, put it in front of users, and see how well they understand it — but not hard-code how the software’s going to present. You can’t change how Congress or other oversight bodies behave, but let’s imagine you have to start another modernization project elsewhere in the federal government. What’s your laundry list of steps to take to avoid somebody else having to clean up your mess later? If a department is doing a software project, they need a cadre of people who are technical experts, know how to manage vendors, and understand architecture. That isn’t always the case. That may be figuring out a way to attract more of these people. It may be the ability to pay them more. USDS had been a really useful version of this. But unfortunately their presence has been significantly reduced. Although you’ve got a different version of USDS, in Tech Force. I’m not going to say it’s a rebrand, but in a lot of ways it looks an awful lot like the old USDS. USDS had a great track record — we’ll see. One thing is, when do you make public commitments? Could there have been some process about what a timeframe that would’ve made sense was? Again, they didn’t have the technical people. A lot is around vendor management and choosing vendors. That whole process has to be reinvented. You were the epitome of one model of going into the federal government: in and out very quickly. You came in on an IPA, you were there six months, your salary was paid by your private-sector employer, and you were out. What are the strengths and weaknesses of that model, versus trying to build long-term technical capacity in the federal government? I definitely think this model is only [to be] used in an emergency. It is much better to have the talent in the government. There are a ton of people in government with these skills, and even USDS, or this new effort — they’re brought in for a limited period. It’s really valuable. But I wouldn’t want those people running things. The nice thing with USDS is they intentionally didn’t try to run stuff, because they knew they were temporary. You need those skills in the government. You need those people driving it. This is a measure of last resort. There’s so much stuff we didn’t know that would’ve been useful. To Aaron’s credit, and Chris Cummings — there’s a lot of people that we were working with that stayed. They know so much more now than we knew. They’re getting so much done, thanks to that knowledge. I much prefer figuring out enough talent that can lead, then utilizing consultants as needed, but not to drive it. This has been a pleasure. If there’s anything I can do to improve my almost-three-year-old’s chance of acing the SAT in almost fifteen years, shoot me a line. I’ll send you the test.