What DOGE gets wrong about tech and government

Lessons learned by civic tech, forgotten by DOGE

(Illustration by David Plunkert/For The Washington Post)

Members of the DOGE network rarely offer thoughtful accounts that depart from the Musk narrative (government is broken, full of talentless hacks, DOGE is fixing things). So I was interested to see a DOGEr express genuine circumspection. This came from Sahil Lavingia, a startup founder, who was interviewed by Ernie Smith about his experience with DOGE.

Lavignia left his start-up, Gumroad, to join Veteran’s Affairs. Here is the key passage:

Now that he’s there, he says he finds himself surrounded by people who “love their jobs,” who came to the government with a sense of mission driving their work. “In a sense, that makes the DOGE agenda a little bit more complicated, because if half the government took [a buyout offer], then we wouldn’t have to do much more,” he says, implying software can replace departing employees. “We’d just basically use software to plug holes. But that’s not what’s happening.”…when it comes down to it, what he’s found is a machine that largely functions, though it doesn’t make decisions as fast as a startup might. “I would say the culture shock is mostly a lot of meetings, not a lot of decisions,” he says. “But honestly, it’s kind of fine—because the government works. It’s not as inefficient as I was expecting, to be honest. I was hoping for more easy wins.”

For anyone with a modicum of experience studying or working in government, nothing Lavignia discovered is novel. Public servants care about public service! There is not really that much waste in government! There are too many meetings and decisionmaking is too slow! Indeed, sir! All true!

To his credit, Lavignia was willing to acknowledge his prior beliefs were wrong. That is a hard thing for any human to do, and something I have not seen from any other DOGE official, certainly not Musk. The exception offers insight to the rule: DOGErs simply don’t understand the government they are destroying, and are unwilling to learn.

For his troubles, Lavingia was fired from DOGE shortly after his interview was published. DOGE employees can have massive conflicts interest, traffic in false conspiracy theories, post racist vitriol online or threaten public employees and be fine. The one thing they cannot do? Admit that government is working better than they expected.

As Elon Musk and other senior members announce their departure from DOGE, it is worth offering a provisional assessment of some key mistakes it has made.

DOGE could have learned a lot by just talking to public employees

This sounds obvious to say, but people handed extraordinary power that affects the lives of others should know what they are doing. If they don’t know important stuff, they should want to learn it before exercising consequential decisions. But DOGE has failed to meet even these banal standards.

A trademark of DOGE was the toxic combination of arrogance and ignorance. Some DOGE members have real and impressive achievements, some do not. But all of those achievements are in the private sector. They know almost nothing about government except conspiracy theories from the internet, or negative interactions with the regulators who oversee their businesses. They did not understand where government spent its money, and that it was not full of waste.

They could have talked to career public employees or nonpartisan government experts. But they didn’t because they did not want to know. Their closest advisors about how government worked were partisan ideologues who want to upend the constitution, Stephen Miller and Russ Vought. Random X posters seem to have more input on the fate of USAID than any policy experts.

This arrogance has had real consequences, including half-assed administrative changes. DOGE ignored their own pick to run Social Security when he told them that their fraud numbers were wrong. They insisted on administrative changes to detect fraud in SSA phone calls. As Natalie Alms reports, the result was that they found two cases of probably fraudulent phone calls out of 110,000. (Or a fraud rate of 0.0018%). But the extra checks they put in place to chase the phantom fraud were not costless: they slowed the processing of retirement claims by 25%.

Why did SSA employees not resist the change? According to a former senior SSA official: “People lacked the fortitude to tell DOGE there was no fraud because they were afraid to lose their jobs. They knew there was no fraud.” It’s not just that DOGE was unwilling to listen: they would fire anyone willing to speak up.

A missed opportunity

A lot of people really wanted DOGE to succeed, or more specifically, they wanted the idealized version of DOGE — smart tech folks disrupting bureaucracy — to succeed. This includes me: when I warned that DOGE would be a disaster right before the Trump inauguration, I felt real ambivalence. Was this just my natural cynicism? Just a couple of days later Will Oremus in the Washington Post asked me, Jen Pahlka and Bridget Dooling to make the case for the best and worst case scenarios for DOGE. Needless to say, the worst case scenarios proved more accurate, and even understated the case, but it was not hard to imagine how things might go well. My version of the best case scenario was:

You get an influx of new talent from Silicon Valley, but you also get the political visibility and momentum that Musk can bring, and you go and fix some problems with how government digital services operate.

Well, we did not get that version of DOGE, but instead the the version of DOGE unable to build a better government because they do not understand what government is, or perceive its value. This is not a matter of talent, but of ideological blindspots.

Some of these blindspots reflect managerial practices that work differently in Silicon Valley relative to government (or almost anywhere else). Henry Farrell makes a compelling case that Silicon Valley leaders are committed not just to “move fast and break things” but to “blitzscaling” — massive and rapid action. This strategy might make sense when firms are in a competition to become the dominant actor in a industry that fuels the creation of private monopolies and untold riches. It makes less sense when you are already a public monopoly, and are subject to capacity constraints and real public accountability. Here, errors are not part of the cost of taking control, but have significant consequences for the public, and engender long term organizational damage. For example, the new SSA Commissioner is reportedly alarmed by the mess he has inherited from DOGE.

There are lots of things DOGE does not get about how to manage in the public sector. This overview of the senior talent that DOGE has pushed out of government is simply an extraordinary indictment of it’s incompetence. But there is one particular type of knowledge that DOGE really missed out on which is unforgivable: how tech works in government.

Some of those who wanted DOGE to succeed were government technologists, frustrated at the slow pace of change they saw during their time in government. Many were ready to welcome fellow technologists with the influence to transform government. But they were isolated, ignored, fired or resigned. In doing so, DOGE missed their best opportunity to actually live up to their brand as tech disruptors who could fix things. Now, their brand will be the tech bros who broke government.

The first administrator of the US Digital Service, Mikey Dickerson sums up this mixed view of DOGE:

I'd say it's remarkable how similar the DOGE and USDS experiences were, considering how dissimilar they were to start. Now I expect to read a lot of "I told you so" and "Good riddance" from people whose careers depend on the status quo. But the fact that the government is so dysfunctional that not even Elon Musk can make a difference is not something to be proud of. I'd like to hope that it drives interest in a real reform effort that is aimed at the core problems in hiring and procurement. His whole experience was 100% predictable and it is what any USDS veteran would have told him would happen. At the same time, I think we have to throw in outsiders and new people that don't know what's 'impossible,' because there is a real learned helplessness problem. The careerists usually don't even move the obstacles that are movable.

The civic tech movement in the federal government was a decade old when DOGE arrived (the US Digital Service and 18F were created in 2014). I have been tracking the evolution of civic tech, because I think it represents one path for the future of American government. That path could have merged with DOGE, creating a single broad vision of tech-driven change in government. Instead, in now offers a competing vision. I think its worth unpacking some of those differences in a bit more detail.

Caveats: Any attempt to characterize two broad movements (civic tech and DOGE) will be replete with exceptions and over-generalizations. I won’t provide a history of civic tech here, which would be it’s own post, and the history itself would be contested. I wrote a working paper that tried to convey civic tech’s role in American government. Basically, think of a bunch of smart technologists (including engineers and designers) who joined government out of a desire to fix stuff. They learned a bunch of stuff on the way, with the best representations of those lessons being Jennifer Pahlka’s Recoding America, or Cyd Harrell’s A Civic Technologist’s Practice Guide.

Lesson: Transparency is not enough

One mistake that defined the early evolution of the civic tech movement was that transparency about data and code would solve problems. Musk has his own version of this claim, which is that he is exposing fraud with transparency and documenting his savings. But in practice, his evidence is so fraught with error and lies as to make the claim worthless. He invokes transparency, while leading the most opaque internal assault on government we have ever seen, where communication occurs largely on Signaland firing Freedom of Information Act officers.

For tech folks new to government, the concept of transparency is very attractive. If I wanted to be mean, I would say that they sometimes assume a lack of transparency to excuse their own lack of knowledge. In reality, there is more information about government programs than any human being can ever read. Again, lets go back to Lavingia:

In contrast to DOGE’s shadowy reputation, Lavingia has made a case for transparency. Based on a pledge he received from Musk during a meeting, Lavingia has been open sourcing his VA work, creating tools that can generate org charts and detect compliance with the president’s executive orders.

If you aren’t a fan of DOGE’s work, the open-source code, while useful for transparency, probably won’t make you feel any better about Lavingia’s work at the VA. The compliance code, for example, is effectively a Python script that hooks into OpenAI servers hosted on Microsoft Azure, detecting whether a federal agency’s communication references chief diversity officers, pronouns, or WHO (World Health Organization) partnerships.

However, Lavingia makes it clear that DOGE has limits, especially thanks to the court decisions and palace intrigue that have removed much of its bite. Ultimately, he argues, it has become a way for roving engineers to get an up-close view of how government works—a “McKinsey for the government,” as he puts it.

A McKinsey for the government. What a concept!

Of course DOGE actually fired in-house consultants like 18F and US Digital Service who were playing that role. For example, one thing Lavingia is probably not aware of is that 18F pioneered open source coding within the federal government. And DOGE was too dumb or ignorant to tap into that expertise, and so they will make the same mistakes that the civic tech movement made, while not taking on board the lessons that civic tech learned.

Transparency is good engineering practice, and a good normative value. But it does not, by itself, solve tech problems. The early theory of change of the civic tech movement was that open data and open-source code would trigger change. This meant that they could work as outsiders and volunteers. But they eventually realized that this theory of change did not work, and so pivoted to directly joining government.

Here are quotes from two leaders of the civic tech movement I interviewed before DOGE was on the scene:

Open-source and open data had a lot of resonance. There was this idea, especially in early Code for America to say, you build something in a fellowship in one city, and then you’d open-source it, and that everybody would just be able to use it…And that reuse theory really didn’t work as well as people hoped.

Maybe 10 years ago, 15 years ago, in the kind of open-data movement days were very much like…sunlight will be the best disinfectant. And I think some of the things that the civic tech movement has actually learned is that it is not…There’s actually work to be done internally.

One reason that transparency did not solve problems is that government employees don’t want code or data: they want solutions that work and fit into the constraints of their world. Government needs permanent internal tech capacity, not short termers who build some tools and walk away (as much of DOGE will in the coming months). It needs people who can build, integrate and manage these tools in response to feedback in a public setting.

Of course, this seems obvious. So why might DOGErs not get it? Why did the early versions of civic tech not get it? Some people I interviewed noted the hacker/libertarian aspect of this worldview: “For folks who are ideologically open-source aligned, the importance of having no gods and no masters is high.”

But I also think the emphasis on transparency reflects a self-serving worldview that government officials are not that bright, and that public sector work is beneath the really smart people, who should be able to just come in, fix shit, and then return to their more lucrative or meaningful opportunities in the private sector. That is a very appealing narrative that elevates the status of the individual holding it, and excuses them from public sector work that is often tedious, bureaucratic and difficult, demanding some skills that they simply don’t have, and investments in expertise that they do not want to make because it will not pay off in the private sector.

Ultimately, those in civic tech who committed to working towards change in government over the long haul, and worked with public officials, found that many were dedicated and knowledgable, and that making change in a democracy requires more patience and humility than building a start-up.

Lesson: Working in government is harder than you think; getting things done requires more than tech skills

As civic tech actors spent more time in government, they gained an ability to see how their work could make a difference and a greater sense of purpose about how to manage change. Again, from my pre-DOGE interviews of multiple civic tech leaders emphasized how managing change in the public sector was a skill, one acquired over time.

We have a lot better idea of what it takes to actually make change. How long, how much work, what kind of institutional engagement. It’s grown by orders of magnitude.

We’ve gotten better and more mature around recognizing that it is not enough just to be a really good coder or a really strong software engineer, and that we can plop those people into government and those people will magically solve problems that career people have been working on for decades. But I do think that a lot of the early learning centered around an initial hypothesis that all that was necessary was to get those people into government and somehow turn them loose and magic would happen.

Part of learning how to use power was learning how to work with public employees, and within government systems. One persistent risk was that civic technologists with limited experience of government or a specific policy domain could easily present themselves as saviors, failing to understand why existing processes worked the way they did. One person I interviewed said: as part of “the movement’s gradual shift from operating outside government to inside, it’s become far less solutionist, far more understanding and supportive of public servants, and somewhat more interested in the underlying causes of poor service delivery”.

The key point here is that working within government is itself a skill, one that involves building relationships with stakeholders, and knowing about policy. This is not a skill that DOGE has developed or is interested in developing in a period of months. So what has it done instead?

DOGE’s undemocratic workarounds will fail

There are a couple of big differences between DOGE and civic tech. One is motivational. The civic tech movement is, broadly, driven by prosocial motivations. They believed that government it slow and unwieldy, but that at some basic level it remains a fundamental means to help a lot of people. If you believe that, it makes sense to work at making the structure of government function. It make sense to invest time in building cool products that can be iterated and improved upon over time (Direct File is a good example).

DOGE mostly believes that government is irredeemably broken, wasteful and fraudulent. If you believe that, it makes sense to downsize government as much as possible, and contract out what you cannot. It makes sense not to build cool products (DOGE killed Direct File) but to build AI that cuts the humans out of the process and automate as much as possible. The fact that DOGE claims about fraud have proven to be erroneous should be a huge red flag about how the assumptions that will be embedded into their AI builds will prove to be wrong in ways that could be catastrophic.

The other main difference between DOGE and civic tech was the attitudes to constraints. Both DOGE and civic tech employees would probably agree that government has too many constraints. The civic tech solution was largely to find ways to work with or manage those constraints. This included user guides to “hack your bureaucracy.”

The DOGE approach has been largely to ignore the constraints. The Privacy Act of 1974 that civic tech people complained about? Pretend it does not exist. Pursue the type of data pooling for surveilling in the context of an increasingly aggressively police state.

The problem of Musk, and DOGE is not a problem of technological capabilities enabled by data sharing and AI. Those capabilities are here, present in the private sector as well as public. The problem is a governance one: laws that exist to constrain the use of data are being ignored. Writing new laws will not solve the problem of people ignoring existing laws.

Civic tech folks complained about their limited ability to shape government. They never had a champion like Musk. But they also largely respected hierarchy and lines of control. Some of the turf issues that occur in any government happened. But the idea of the US Digital Service, rather than the Secretary of the Treasury, trying to appoint the acting head of the IRS is unthinkable. But Musk did so. The idea that a Cabinet Secretary would hand over all operational decisions to 18F would have been considered bizarre. But some Cabinet Secretaries have done this with DOGE.

The logic of DOGE’s approach is that they cannot be doing damage by destroying government, because nothing government does have value, and constraints like laws are merely blockers to ignore. And so, we have people who were elected by no-one exert extraordinary power over public services they barely understand, the consequences of which will play out long after they have left government.

There are examples across government, from out scientific agencies to national parks to Social Security or IRS. The one I have never been able to get over is USAID. Here, Musk officials, contrary to the law, contrary to the announced priorities of the Secretary of State to protect life-saving programs, manually switched off those programs. From Matt Bai in the Washington Post:

Rubio had decreed that certain critical programs — such as aid to Ukraine and Syria and costs related to the PEPFAR program to combat HIV in Africa — would continue to be funded. Several times, USAID managers prepared packages of these payments and got the agency’s interim leaders to sign off on them with support from the White House. But each time, using their new gatekeeping powers and clearly acting on orders from Musk or one of his lieutenants, Farritor and Kliger would veto the payments — a process that required them to manually check boxes in the payment system one at a time, the same tedious way you probably pay your bills online. Meanwhile, AIDS clinics shuttered and staff found themselves stranded in unstable countries such as Congo. A pregnant woman in an undisclosed country has sued the Trump administration because she was denied a medevac helicopter. In another case, I was told, an employee in southern Africa who needed chemotherapy was also denied a chopper because no one would authorize the money.

Much of what DOGE is doing is illegal. Much of it will create long-term and lasting damage to America. This is the clearest example of an action that was illegal, anti-democratic, but also sped past any moral grey zones to being deeply evil. Even still, 60,000 tons of food, enough to keep millions from starving, are rotting in warehouses, blocked from release by another DOGEr, Jeremy Lewin. Per an analysis in Nature, up to 25 million people will die because of US cuts to foreign aid. Those who die will be largely the most vulnerable people in their world, their fates decided by some of the most privileged and protected people in the world.

DOGE was able to have an impact not because it brought gifted technologists into government, but because of the broader Trumpian indifference to key aspects of governing, such as capacity, service quality or accountability. They have broken much, and built nothing of consequence.