The federal government is losing the workers who build and oversee its artificial intelligence tools, even as AI use across agencies keeps climbing. That gap is at the center of a new call for a public scorecard to track the results.
The proposal comes from Gene Lockwood-Shabat, a former senior official at the Department of Veterans Affairs, in an opinion piece published November 4. He argued that Washington has no consistent public way to judge whether AI growth and staff cuts are improving government services.
AI Use Rises as Staff Numbers Fall
The Government Accountability Office reported that AI use cases across 11 federal agencies nearly doubled from 2023 to 2024. The count rose from 571 to 1,110.
Generative AI use cases grew even faster over the same period. They climbed from 32 to 282.
The workforce moved in the opposite direction. Across 22 major agencies, nearly 378,000 employees left during 2025, while about 127,000 were hired.
Those agencies had more than 250,000 fewer workers in January 2026 than in December 2024. Lockwood-Shabat said counting jobs cut or AI tools adopted does not show whether services got better.
The Internal Revenue Service offers one example of the staffing risk. IRS officials told GAO that its Research, Applied Analytics and Statistics Division lost 63 employees, most of whom had worked directly on designing, building or overseeing AI projects.
IRS officials also said the agency had not properly identified the AI skills it needed. That raises questions about whether the IRS has the expertise to run the tools it plans to rely on more heavily.
Scorecard Modeled on FITARA Grades
Lockwood-Shabat proposed that Congress and the White House Office of Management and Budget create a unified AI and Workforce Management Scorecard. It would build on the scorecard Congress used for years under the Federal Information Technology Acquisition Reform Act, known as FITARA, which gave agencies public letter grades on IT management.
Under the plan, agencies would report on each major service that uses AI. They would list spending, changes in staffing and skills, performance against existing service targets, and quality data such as AI-related error rates and rework.
Each measure would include a starting baseline and later reporting periods. For example, if an agency says AI sped up application processing, the public could see whether errors changed and whether the gains lasted.
Lockwood-Shabat said the scorecard would not require a new reporting system. Agencies already publish performance plans and budget justifications, and the task would be to link that data in one place.
He added that measures would need to fit each service. An error in a clinical setting, he noted, is not the same as a delayed tax response.
GAO has also recommended that the Defense Department, the Department of Homeland Security, the General Services Administration and the VA update their policies. The change would require agencies to collect lessons learned from AI purchases in a systematic way.
The most recent warning came from the VA. In June 2026, the department’s inspector general found staff were using general-purpose generative AI chat tools for clinical work.
The inspector general said the VA could not measure how widely those tools were used or readily identify AI-related patient-safety events. The VA agreed with the inspector general’s recommendations.
