Fire First, Automate Later?
Here is a government modernization story with uncomfortable timing: the General Services Administration has lost a large share of its workforce and is now pursuing a target of automating or saving a million work hours. That is a bold number to put on a slide when the people left in the room already have plenty to do.
But the sequence needs care. A reduction in headcount is not the same as every departing employee being fired, and a target for hours saved is not evidence that AI has replaced the people who left.
The Million Hours Challenge
Federal News Network’s April 14 report puts GSA’s workforce reduction since October 2024 at nearly 40%. At that day’s OpenText Government Summit, Deputy Administrator Michael Lynch described a million-hour automation goal and roughly 400,000 hours of potential work already identified.
Lynch framed the effort as freeing people from lower-value tasks, rather than replacing them. The report also describes GSA Labs: about 300 employees applied, with an initial group of roughly 30 contributing alongside their existing jobs. That is an initiative under way, not a completed demonstration of a million hours saved.
For scale, a million hours equals about 500 working years if each contains 2,000 hours. It does not follow that the agency can remove 500 specific jobs. Saving fragments of time across many tasks is different from eliminating whole positions, and supervision or correction can absorb part of an apparent saving.
What the Watchdog Actually Found
A separate GAO report released April 14 gives the staffing concern firmer grounding. It found that GSA’s Public Buildings Service had not conducted strategic workforce planning before its staffing fell about 45%, from almost 5,700 in September 2024 to just over 3,100 in November 2025.
The departures included voluntary programs and involuntary reductions. GAO reported resulting skill gaps and concerns from tenant agencies, including delays in obtaining cost estimates needed for property-disposal projects. It recommended better performance measures, stakeholder feedback, monitoring and workforce planning.
That is a substantial criticism of the reorganization. It is not a finding that an AI chatbot caused the staffing problems, nor an assessment proving that the new automation effort can solve them.
USAi Is Not New This April
GSA announced USAi in August 2025 as an AI evaluation suite for federal agencies, including chat, code generation and document summarization. It should not be presented as a chatbot invented this month to fill empty desks.
Nor is an old list of models a reliable description of what is available now. On February 27, GSA announced Anthropic’s removal from USAi. That makes a description built around older Claude versions particularly misleading.
Which information a system can handle also depends on the approved environment and use case. A blanket claim that it cannot touch any non-public information needs evidence specific to the deployment. The practical question is whether the approved tool can access what a task requires while preserving the necessary controls.
The Hard Part Is Measuring the Result
Drafting an email faster is useful. Providing a reliable cost estimate, checking an unusual procurement issue or maintaining a building involves a different mix of knowledge and responsibility. “AI” is too broad a label to tell us which of those problems has actually been solved.
The enormous investment in AI companies makes ambitious promises easy to find. For an agency, the meaningful evidence would be time saved after review, fewer errors and better service for the people relying on it.
GSA’s challenge may produce useful improvements. The documented staffing gaps still need their own response. A slide saying “Eliminate, Optimize, Automate” can organize a presentation; it cannot, by itself, organize a workforce.
Our preferred government efficiency initiative remains shortening meetings. The cat has already automated attendance by sleeping through them.
