Remember Aaron Lemon-Strauss, the executive director of FAFSA who wrote the brilliant response to a GAO report defending the product model? Well, he’s just joined Recoding America as our chief digital and AI officer (hooray!) and has a few more wise and motivating thoughts on this moment and how to meet it. Welcome, Aaron!!
— Jen
When a group of us joined FAFSA in the summer of 2024 it was already on fire. Our very presence was evidence of the crisis: six private sector technologists with exactly zero collective days of government experience showing up and suddenly calling meetings and engaging vendors. America’s largest single investment in the future had broken and everyone agreed it needed to be fixed.
The thing about a crisis is that it creates permission. Paperwork Reduction Act approval, which normally requires four to five months? Turns out there’s a waiver for that. Updating a SORN is a nine month process. Did you say the FAFSA’s on fire? In that case we’ll push the new version live next week. That Computer Matching Agreement (CMA) you’ve been working on that we told you would take eighteen months? Let’s call the head of the right office – ok, it’s all sorted.
As we were making progress on these fronts, teammates kept telling me that the way we were working was setting a bad precedent. If we could get our SORN updates published in weeks instead of months, everyone else who was waiting in line would be pissed off. And how would they explain the divergence in process to GAO? “We know you aren’t here forever, Aaron, and you’re creating a mess and leaving the rest of us to pick up the pieces.”
It’s a fair point (and, sorry! xoxo). But no one ever said, “CMA renewals need to take eighteen months because we have to charter a tall ship and sail it through pirate-infested waters to bring back a treasure trove of rare vibranium that can only be mined during a one-week period over the Lunar New Year.” It takes eighteen months because we let it: because a cascade of rigidity turned a noble policy goal (transparency in interagency data sharing) into an eighteen month bureaucratic maze. My colleagues’ concern is one of internal fairness, not a defense of a process that shouldn’t be defended.
The FAFSA crisis created permission to work better. But the team is right: the way we worked wasn’t sustainable. And it isn’t good policy, either. Getting to run roughshod over process because you happen to be the loudest is no way to run a government.
But the status quo sucks. That’s no way to run a government either. Reformers have been trained to start planning for the next crisis. They prepare plans for how they can change government once the permission window reopens. But this past history of crises tends to sit within specific agencies or programs, which means we never get to refactor the government itself.
Artificial intelligence is different. It presents a crisis to both specific programs and to basic ways of operating. AI creates a crisis by making the assumptions on which government operates untenable.
If your bank, airline, and phone can understand what you’re asking, instantly synthesize information, and resolve increasingly complicated problems, “please fax us this form and expect a response in six weeks” becomes much harder for government to defend. The baseline for what people regard as functional is moving underneath us.
Now, AI isn’t actually there yet. I’m looking at you, Verizon customer service bot. But the cute haikus of 2024 are now the autonomous coding agents of 2026, and the eventual improvements to user experiences feel inevitable and forcing. The market will not allow firms to stand still. Even Verizon.
As this change rolls out across the economy, citizens and institutions will demand more from government, and if government is stuck in amber it will generate more and more headlines.
Here’s a recent example: Federal Student Aid is about to forgive $11 billion in loans because the plan to sort legitimate claims from fraudulent ones involved hiring dozens of attorneys to review each claim manually and they couldn’t get it done within a court-imposed deadline. In the past, government could point to the sheer scale of the manual task as an explanation for why something couldn’t happen faster. Increasingly, that explanation isn’t going to work. In this case, the court was unwilling to treat the Department’s inability to process 170,000 claims through its existing system as a reason to move the deadline. The process couldn’t keep up; the deadline didn’t move. The government wrote $11 billion off of its balance sheet.
Maybe you say, “C’mon, everyone knows that government services are bad, and they accept it. Take your surprisingly delightful FAFSA back to fantasy land. Just look at the DMV.” Alright, let’s look at the DMV.
In 2018 the California DMV was so bad that it became an issue in the governor’s race. Governor Newsom came in and declared that the DMV had “failed in its fundamental mission.” There was a strike force. Another crisis, another set of digital reforms, and satisfaction went from 2.5 out of 5 to 4.1 out of 5.
Same thing is underway in North Carolina. Same public pressure on unreasonable government service delivery, same playbook of overhauling process, same positive results. By May 2026 average wait times at the NC DMV were down nearly 90%.
Crises aren’t defined by how bad a problem is. They’re defined by when the bad outcomes hit people with the ability to make a stink. AI is moving the window of what is acceptable. Things government has long explained as difficult, expensive, or slow are going to look increasingly like choices.
At FAFSA, a broken website made the crisis obvious for us. AI gives government the chance to create that same permission structure before something breaks. The point isn’t to temporarily suspend the rules, as we’ve done in past crises. It’s to ask which rules protect an outcome, which merely protect a process, and what we would build if we stopped assuming the process was inevitable.
So what should government leaders do? My own experience with Secretary McMahon’s leadership aligns well with Colorado’s recent moves. The key first step: bring technical leadership into the agency in positions of authority and give them autonomy to operate. Good news: this isn’t cost-additive, you’re already paying for this, you’re just doing it through contracts with corporations domiciled in Delaware. Move the technical management layer from vendors to the government and you can reduce overall headcount and cost while improving outcomes. Jen will be writing more on this soon.
We shouldn’t wait for the website to catch fire or the hearings to start. This is the crisis we’ve been waiting for.
I like the point of “Move the technical management layer from vendors to the government.” To me it seems like there’s this thing that happens in government where some function gets outsourced, it sucks, and then nobody in the government takes any responsibility for it sucking, and it doesn’t even seem like they understand what’s going on or why. Contracting out can be great but you need enough technical ability on staff to be able to take responsibility for the performance of the contractor.
Great thoughts here, Aaron, especially, "It’s to ask which rules protect an outcome, which merely protect a process..." I would extend this a bit to ask, "Which merely protect a job?"
I've seen many deployments of AI/ML over the last 13 years. What's become evident with the latest GenAI cycle, at least in my experience, is actually something you also said: "Things government has long explained as difficult, expensive, or slow are going to look increasingly like choices." Yes! Same goes for private companies, where many times a long-standing middle management layer has not really done much in a decade. Not much to innovate, move things along, or improve-- just attend meetings, set a policy, etc.
AI can expose where inefficiencies not only existed, but that the inefficiency was intentional. As you said, you take something from months to weeks or days, and everyone upstream freaks out because, now, THEY need to move faster and do more. Well the first step to treating a wound is exposing it to oxygen, then clean, then repair and bandage. We're at the expose stage, which is much needed.