What DOGE taught us about AI and federal workers

Last year, the U.S. Agency for International Development lost 97% of its staff in a matter of weeks. An article published in The New York Times last month found the majority of these former employees were still out of work a year later — not between jobs, but out of the market entirely, with some managers who once earned six-figure salaries applying for part-time retail positions.

I watched this happen. I worked at the State Department until August 2025 and helped create a pro bono coaching network for impacted colleagues, many of whom were deeply traumatized. After thousands of hours of those conversations, one question kept surfacing: who am I now?

The cuts were political, not technological. But strip that away and what remains is the most concrete demonstration we have of what happens when a large category of federal professional work disappears faster than any system can absorb it — and why the standard policy response is not enough to cover these numbers. Over 270,000 federal employees separated from the U.S. government through layoffs, forced resignations and buyouts. 

Most impacted workers did not lack skills. They lacked a place where those skills made the most sense. Federal workers who spent careers running HIV programs or managing humanitarian operations did not simply need to update their LinkedIn profiles. They lost the institutional context that made their expertise meaningful. The formal policy response was minimal. Workers relied on informal networks. The DC labor market, despite being one of the most credentialed in the country, has not absorbed their talent.

This is where the federal story becomes a story about artificial intelligence. As deferred resignation agreements were being signed in August 2025, the U.S. government licensed ChatGPT to all federal agencies for a dollar. The State Department reframed AI as the vehicle for development outcomes that USAID's human expertise previously delivered. OpenAI began offering grants to NGOs in regions where USAID once operated. 

This is not a Washington anomaly. The Economist devoted its cover earlier this month to a question that was once considered alarmist: whether AI could produce the most significant disruption to working life in a generation. The answer, even among economists who were recently skeptical, is increasingly, possibly ‘yes’ — and governments should not wait to find out.

New data from the Bureau of Labor Statistics show AI-exposed occupations are already losing jobs, and government employees could be among the most vulnerable, given the large concentration of workers handling the analytical, administrative and policy roles where AI capabilities are advancing fastest. The official numbers are not catching up fast enough. By the time they do, the adjustment will already have failed for the workers caught in the first wave.

There are three things federal agencies and policymakers should be doing now: First, plan for the fiscal squeeze. Federal workforce costs are not just a spending question — they are a revenue question. As AI shifts work from human labor to automated systems, the income tax base that funds agencies, benefits and services erodes at exactly the moment demand for support rises. Brooking Institution modeling shows this fiscal pressure could be severe. Agencies need fiscal scenario planning now, not after the trend is visible in budget projections.

Second, design workforce transition for what people lose. The Office of Personnel Management’s current transition support is built for skills retraining. The evidence from the DOGE displacement, and from every serious study of mass professional job loss, is that the harder problem is purpose and identity, not capability. Transition programs that ignore this will produce the same frustration the DOGE coaching networks documented.

Third, ensure that AI deployment decisions in federal agencies are not made solely by the vendors supplying the technology. The informal support networks that emerged in Washington show what community-level resilience looks like when institutions fail. They deserve federal attention and funding, not just admiration. Workers, communities and agencies affected by AI deployment decisions need a meaningful voice in how those decisions are made.

The DOGE cuts were political, but what they demonstrated is not. Federal agencies are the first institutions in America to run this experiment at scale. The question is whether anyone in government is paying attention to the results.

Kristen Cordell served as Senior Advisor at the US Department of State until August 2025 and is currently Senior Director of Policy at Grand Challenges Canada. Adrian Brown is Founder and CEO of Windfall Trust, a nonprofit working with governments and policymakers on AI economic preparedness.