GSA Lost 40% of Its Staff. Now It Is Deploying AI Agents.
GSA lost 40% of its workforce since October 2024 and is now running an AI platform to automate a million hours of work. That gap between the 82% adoption headline and the agencies filling it is the real story.
The General Services Administration has lost nearly 40% of its total workforce since October 2024. Its response: launch an AI platform and challenge its remaining staff to automate a million hours of work. That is not a digital transformation. That is triage.
When I queried Agent Kyle, Kyle, Texas's 311 customer service agent and one of the first deployed government AI systems, I got a clean handoff for pothole complaints and streetlight outages. The routing worked. What it could not tell me was who reviews the cases it routes incorrectly, or what the escalation path looks like when the system fails. That gap is the actual story of government AI adoption right now.
A Salesforce-commissioned IDC study making the rounds this week says 82% of government organizations have adopted AI agents, a number several outlets have run without scrutiny. The survey polled 118 self-reporting government decision-makers in March 2026 and counts any pilot as adoption. KPMG's Q3 2025 pulse survey, using a more conservative deployed-systems threshold, shows 54% of private sector organizations have integrated AI agents into operations. The methodologies are different. The comparison is also not the story.
The SSA's public AI inventory CSV is more revealing than the survey. It lists classical machine learning tools for disability benefits adjudication alongside a structured pipeline of pilot programs. The GSA launched USAi in August 2025, the government's first shared platform for AI experimentation, and is now running a million-hours challenge to automate 400,000 hours of administrative work. Michael Lynch, a former SpaceX executive who became GSA Deputy Administrator, is running it. This is not a career civil servant managing a technology modernization project. It is someone from the private sector running an automation campaign inside a workforce already cut by two-fifths.
The accountability question that follows government AI becomes more acute here. DOGE-era workforce cuts eliminated not just the people but the residual capacity to notice when the system makes a bad call. Who catches the error? The IDC survey does not ask this. Neither does the 85% productivity claim, which represents government leaders estimating AI saves their workers up to 45% of their time per week, self-reported and unverified. KPMG's data shows 65% of leaders across sectors report difficulty scaling AI use cases to deliver measurable return. The government survey did not ask about scale failure.
None of this means the government AI story is vapor. The SSA inventory is real. USAi is real infrastructure. Agent Kyle routes real complaints to real departments, and it works for the use cases it was built for. When the EPA and IRS announce plans to rebuild capacity through automation after deep staffing cuts, those are real deployments, not press releases.
But real and ready-for-oversight are different things. The 82% adoption headline says government is ahead. The 40% workforce reduction says government is running a different experiment: one where the humans who would normally catch AI errors have already been let go. The number to watch is not the survey stat. It is whether the agencies that cut deepest are the same ones now betting hardest on automation to fill the gap.
