My first stint as a government data scientist went to some places that, at the time, I would have described as ‘dramatic.’

For instance: a coworker at the same level as me emailed me an org chart showing that I now reported to her. She told me I couldn’t share it with anyone. Naturally, I did. Turns out I did not, in fact, report to her.

But the drama wasn’t even the biggest issue. The real problem—and this has been the case for every technical job I’ve done in government—was figuring out, between me and management, how to identify projects that were appropriate for my skill set, possible to do given our team’s technical and personnel constraints, and useful to my agency.

On paper, this should be simple: there are technical problems, you bring in people with the skills to solve them, voila. In practice, it’s not. Which is how I’ve found myself, at various points, wrangling spreadsheets instead of coding, managing contractors, and building models that were never going to get used.

At my most recent civil service job at the short-lived AI Corps at the Department of Homeland Security, this was done well—but I didn’t realize at the time how much work was happening behind the scenes to enable that. We had leadership with previous experience with digital services teams in government, which meant we had a significant advantage in doing it right. And it was still a challenge.

And that was before everything went sideways.

The federal government ended 2025 down roughly 250,000 people. Technical folks were not excluded from that number. 18F got dismantled—that was the team that built tools like cloud.gov. The U.S. Digital Service (USDS) got taken over by DOGE, fired some people, and saw significant resignations. Some agencies conducted terminations of probationary employees—anyone hired or promoted in the past year or two.

At Homeland Security, some engineers were fired for being probationary. More left semi-voluntarily, through a combination of early retirement and buyouts as the carrot, and return to office (including for employees hired to be fully remote, which included everyone on AI Corps), forced reassignments requiring relocation, and a fear of additional layoffs as the stick.

I loved the work at AI Corps. Then the chaos started. Projects stopped. We were told by leadership that there weren’t plans to eliminate our team, but also that individuals likely would be fired—and the indications were all negative that things would be functional again.

And then came the Sunday afternoon email ordering us to report back to work five days a week or be fired starting the next morning.

I left shortly after that.

Last month, the Office of Personnel Management (OPM) announced several new programs: U.S. Tech Force (1,000 fellows annually), Data Science Fellows (250), and Project Management Fellows (250). These are all short-term roles where OPM assists with the hiring, but actual assignments are to various agencies across the executive branch.

The goal? “Modernize the federal government” and “tackle the government’s most complex and large-scale challenges.”

These are good goals. If you want to work on them, I hope you find a role where you can do that. And I also hope these programs are successful.

But they might not be. So if you’re going to apply, go in with your eyes open about what to look for.

Government tech modernization is difficult. OPM spent approximately $290 million on a project to automate federal employee retirement claims before canceling it in 2011. The Coast Guard spent $67 million over five years trying to replace its aging health records system before scrapping the effort; as of 2018 it was still paper-based. The VA spent $127 million over nine years on ascheduling system that never launched. The FAA’s air traffic control systems include multiple systems over 30 years old; its ongoing modernization effort, NextGen, has faced decades of delays, and some critical systems won’t be modernized until 2035.

I think of the core problem as responsibility without authority.

For instance, it’s your job to modernize the agency’s data systems. But you’re reliant on contractors to do much of the work because you’re not staffed to build everything the agency wants fixed by yourselves. The contracts have already been competed and awarded, most likely to the vendor putting forth the Lowest Priced Technically Acceptable proposal, not the best. The payment structure also likely incentivizes billing hours rather than building good products.

Additionally, the roles and skill sets the contractors were hired for may not be the ones you need and the work may not be scoped in a way that’s useful to you. And if it’s not working out, you can’t just change the contract or bring in different people. That takes months or years, if you can do it at all. (And you probably can’t, because you don’t know how and your office doesn’t even own the contracts.) And good luck if you try to get involved in pushing the contractor to hire or screen differently.

Or maybe the contract is fine, but you need existing code from the previous contractor or else you’re way behind. Where does the code live? Not on any server you’re going to get access to, because whoever was managing that previous contract never asked for it.

You also can’t change hiring regulations for civil servants. You may not even be able to get HR to do things that are totally legal.

And then there’s risk aversion about the wrong things. A lot of people are terrified of a process violation, of something that could look bad in an audit. But they’re not as worried about the system being terrible and not working for users, or wasting years and millions of dollars on something that doesn’t solve the problem.

Meanwhile, your leadership is under pressure for short-term wins. They just brought in a bunch of expensive, highly visible staff. They need to show results. But they may not know where the opportunities to be effective are—what’s realistic to accomplish, what would require years of groundwork, what’s blocked by forces outside their control, like Congress. They may want innovation but settle for innovation theater.

Finally, a lot of people at the agency you work in and whose assistance you need are predisposed to dislike you. You’re likely younger, you’re paid more, and your mission (to fix the bad stuff) is an implicit criticism of them. I’m not saying everyone sees it that way, but “they thought they were the smartest people in the room and they didn’t listen to us” is something that gets said so regularly about digital services teams in government that it’s hard to tell when it’s true and when it’s not1.

The limitation on making progress in government is rarely engineering skills. Building the thing is usually the easiest part. It’s everything else that’s hard.

And all of that is just the starting point, in a good year2.

Think about the year civil servants—the people you will be working with and need buy-in from3, should you do any of these programs—just completed.

Russell Vought, who ran the Office of Management and Budget in Trump’s first term, said in a private 2023 speech that in a second Trump administration, “We want the bureaucrats to be traumatically affected. When they wake up in the morning, we want them to not want to go to work because they are increasingly viewed as the villains.”

“We want to put them in trauma,” he said.

And that’s pretty much what happened in 2025. People were told with no notice to return to offices with no space for them. Folks you’d helped recruit and persuade to join the government were fired, or left because they had good reason to think they would be. Staff were called into meetings with DOGE representatives who wouldn’t turn their cameras on and asked to justify their jobs and offer judgment on their coworkers. Hit pieces were written calling out specific employees in ways that appeared coordinated with the people making personnel cuts.

And all of that was just a small subset of the things that happened to people I know personally.

So when you show up as a new Tech Force fellow, there’s going to be additional skepticism besides just the normal kind. People just watched their colleagues with a lot more experience in your fields get pushed out or fired. And here you come, early career, only here for a short time, and hired to do...what exactly?

Also, it’s not as if everything has settled down: you’ll either be subject to ongoing actions or you’ll be insulated from them. Either is a problem.

And there’s another layer of instability: contract actions. Last year saw widespread cancellations of federal contracts and delays in awards and extensions. At Homeland Security, contract actions of $100,000 or above still require sign-off from the Secretary. You will not be directly affected, but the contractors you work side-by-side with and depend on? They may be.

Finally, if the starting assumption from the agency leadership that brought you in—and potentially even from your managers brought in from industry to manage you—is that the government doesn’t work because civil servants are stupid or lazy, you’re starting from a terrible position. You need to build trust with people to get things done. That’s hard when the message from above is contempt and when you’re under pressure to put out a narrative, both in the agency and externally, that minimizes the contributions of career civil servants.

Here’s what to look for when you apply to reduce the likelihood that you’re walking into a situation like that.

When you’re interviewing, look to see that someone running this program at your agency has an actual theory of change about building things in government and a history of successfully doing so. Get names and find out what their backgrounds are.

Ask questions like:

What conversations have happened with agency leadership? How have they prepared for us?

Have projects been scoped yet? What does that process look like?

Regarding those projects, what’s been tried before at this agency and what are we doing differently? Who is currently working on them and how are we going to work with them?

What infrastructure exists for us? Is there a git server? Are we going to be allowed to use AI coding tools, and if so, what?

What’s the process for user research? Are we hiring or working with UX researchers or designers? How will we talk to the people who use these systems?

Do not just assume this has been figured out. I’m looking at OPM’s long list of participating agencies and I would be surprised if when that was released, they’d had extensive conversations with each one to assess that they’re going to make productive use of you.

I’m sure agency leadership said “by all means, send us technical people”, but did it go beyond that, or will it before they start hiring? And if the agencies don’t have good plans, will OPM back down from their very ambitious, very public hiring goals?

Normally, this could get deferred slightly, and they could work things out after you show up. But these are one and two-year terms. If there aren’t projects scoped by the time you get there, you could easily burn through a big chunk of your term waiting for something useful to do—or discover that the people you’re working for don’t know how to find that.

My prediction: Experiences will vary wildly agency to agency. Some agencies will do this well and others will not.

A lot of tech within the federal government is bad, and the challenges to improvement aren’t inherently insurmountable. Even last year, good things were built.

And I don’t think the premise or structure of these programs is wrong. Centralized hiring is absolutely the way to go, because getting hiring announcements out and screening resumes is a huge lift that most agencies don’t have the capacity to do. (Does OPM? They can staff up to do it, and it’s easier to solve this for one agency than for all of them.)

I also think it’s good that they’re making these positions explicitly short-term. No one should be joining the government now planning on having job security. And it would not surprise me if there are problems that get solved faster than they would have previously—for instance, I think if you join, you’ll quickly get the computers you need with the software you need, which isn’t nothing.

Finally, making these programs high-profile, working with industry, targeting folks early enough in their career that they can be paid very competitively on the GS pay scale—this is all fine.

But if your goal with joining these programs is solving actual problems for actual people—and I hope it is, if you’re thinking of applying—there are also significant risks here, and reasons to be skeptical. The usual structural challenges of government tech still exist, compounded by everything else that’s happened.

If you apply, go in with your eyes open. Ask hard questions. Look into the backgrounds of folks running these programs at the agency you’d be assigned to. Don’t assume it’s going to work out. And understand that you’re not just signing up for a technical challenge; you’re signing up for a political and organizational one too.

1

It is, in fact, true sometimes.

2

If you want deeper context on the structural challenges, read Jennifer Pahlka’s Recoding America

3

Why do you need buy-in at all? In large bureaucracies, goodwill is incredibly important. That’s for a few different reasons. One is that other people have the authority you lack and need. Another is that you don’t know what you don’t know, and that’s compounded if you’re new to the government or even just to that agency. For both reasons, you’re likely to be reliant on other people to go out of their way to assist you, even if it’s not exactly their job to do so, and may in fact take them away from their job.

No posts