Across three consecutive administrations, the federal government has worked to expand the use of artificial intelligence across agencies.

Regardless of who has held the White House, the pitch has, for the most part, remained consistent. When used with appropriate caution, AI can speed up service delivery, sharpen decision-making, strengthen national competitiveness, and bolster national security.

In a recent report, I found that push to adopt AI in federal agencies has led to a significant increase in reported use.

So far, however, AI adoption is concentrated in a handful of agencies and continues to confront structural constraints that have hindered technology modernization broadly and pose new challenges due to the dynamic, uncertain, and sometimes inscrutable trajectory of AI development.

Increased agency AI use

In 2023, agencies documented around 700 AI use cases. By 2025, that figure had climbed to more than 3,600—a fivefold increase in two years. In the most recent inventory released by the Office of Management and Budget (OMB), forty-one agencies submitted AI uses, up from 21 two years earlier.

Some of this growth is undoubtedly due to clarifications in guidelines by OMB and agencies changing how they report use cases, but that does not wholly explain the rapid—and increasingly sophisticated—experimentation reported by agencies. AI is helping to facilitate routine back-office tasks, but it is also increasingly being embedded into mission-enabling workflows tied to benefits processing, medical service delivery, and law enforcement.

This aggregate growth, however, conceals disparities across agencies. Since the documentation effort began three years ago, five large agencies have accounted for more than half of all reported use cases. In 2025, the average large agency reported 211 deployments, up from 114 in 2024.

By contrast, the average midsize agency reported 48 uses and the average small agency reported just five, up from 32 and four respectively in 2024. Some of this divergence reflects different missions and risk profiles, but it may also reflect the fact that experimentation requires capacity and resources, which are unevenly distributed.

Finding and retaining the right talent represents one persistent bottleneck. According to federal jobs data, the government has posted more than 56,000 technical job listings since 2016, but fewer than three percent explicitly called for AI capabilities.

A hiring surge tied to the Biden administration's 2023 executive order pushed AI-specific postings to roughly eight percent of technical jobs by 2024, with about a third using expedited hiring authorities (as compared to 1/5th for other technical jobs).

This nascent, but targeted expansion may have been interrupted by workforce reductions early in the second Trump administration. When technical talent does enter agencies, however, limited prospects for career advancement make them difficult to retain. Term limited appointments have helped to fill some of these gaps, but technologists are limited in what they can accomplish in just a few years.

Technologists also operate within a culture that historically rewards caution, which makes innovation and experimentation more challenging. Several technologists I interviewed emphasized that successful AI pilots depended heavily on whether senior leaders gave explicit permission to experiment.

Where that permission is absent—because leaders lack funding, technical understanding, or have more urgent priorities—even capable staff tend to retreat to more conventional approaches.

Unpredictable forecasting capabilities

Procurement rules and budgeting processes compound the difficulty. Federal budgeting cycles begin a year and a half before the start of a fiscal year, which requires agencies to forecast capabilities for a technology whose trajectory remains unpredictable. Authorization frameworks like FedRAMP, authorizations to operate, and the Federal Acquisition Regulation were designed for relatively static software with predictable updates.

The Paperwork Reduction Act, which can require six to nine months of OMB review for data gathering activities, makes it difficult for AI systems to learn from user feedback. Although they are well-intentioned, these processes may create bottlenecks that slow adoption. Efforts to speed up procurement and make pockets of funding more flexible do exist, but they still struggle to keep pace with AI developments.

Widespread public skepticism of AI adds another complicating layer that hinders the adoption of AI-enabled solutions. Pew Research Center data shows that about half of Americans report being more concerned than excited about AI's growing role in society, up from roughly 37 percent four years earlier.

Only 17 percent expect AI to have a positive impact on the country over the next two decades. The absence of comprehensive federal AI legislation, rhetoric around job displacement and potential AI catastrophe, and the increasingly politicized nature of AI procurement have not made the case for embedded AI into government operations an easier one to make.

Improving federal adoption of AI

Improving federal adoption of AI will require sustained investment in technical talent, procedural and budging reform, and transparency to build trust with the American public.

For talent, clarifying the purpose of fixed-term programs like the new U.S. Tech Force, building genuine career ladders for technologists, developing more shared resources, and treating AI literacy as a core competency rewarded in performance evaluations will help facilitate AI adoption across government.

Revisiting acquisition rules, authorization requirements, and budgeting cycles to build in more flexibility for iterative and uncertain technological trajectories will help to speed up bureaucratic processes that can be cumbersome.

To foster and maintain trust, it will be important to improve on consolidated use-case inventories, fully document risk mitigation for high-impact systems, and prioritize visible, beneficial deployments of AI enabled systems that demonstrate value to citizens, for example in the areas of tax assistance, benefits navigation, weather prediction, or outbreak surveillance.

Research shows that satisfaction with public services is a significant driver of trust in government. With public trust in Washington near a historic low, AI deployments that improve how citizens interact with their government could, over time, contribute to rebuilding that trust.

However, cautionary examples from abroad, including automated benefits scandals in the Netherlands and Australia, loom over any large-scale federal deployment. Visible failures could just as easily erode trust even further.

The bipartisan continuity in federal policy suggests broad agreement that AI has the potential to transform how government serves its people. But to do so effectively will require a federal workforce, procurement system, and public communication apparatus designed to keep pace with the uncertainty and unease brought by a technology that is evolving rapidly, and sometimes, unpredictably.

We've featured the best IT Automation software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Fellow, Foreign Policy, Artificial Intelligence and Emerging Technology Initiative, Brookings Institution.