AI reprices public-sector knowledge work

Artificial intelligence is no longer just a technology story. For government, it is becoming a budget story, a workforce story and a procurement story all at once.

The sharpest signal comes from the market, not the hype. In a recent paper by Ryan Stevens of Ramp, analysis of firm-level payments data found that among the most exposed firms, each $1 drop in spending on online labor marketplaces corresponded to roughly three cents in added AI model spending by the third quarter of 2025. That is a startling ratio. It suggests that for a meaningful slice of routine knowledge work, buyers are finding a way to get acceptable first-pass output at a fraction of the prior cost.

Federal agencies should resist the temptation to read that number as a simple substitution story. Government does not buy work the way startups do, and public-sector value cannot be reduced to token costs. Still, the ratio matters. It signals that generative AI is changing the economics of drafting, summarizing, coding assistance, document review and other language-heavy tasks. For public managers facing hiring constraints, contractor scrutiny and pressure to deliver faster, that shift is too large to ignore.

What the 25-to-1 ratio means in government

In government, a 25-to-1 cost gap does not mean “replace people with chatbots.” It means managers should examine where expensive human effort is being consumed by low-risk, repeatable, text-based work. That includes first drafts of correspondence, brief summaries of long records, initial research scans, meeting notes, translation support, code scaffolding and internal knowledge retrieval.

The federal policy environment already assumes agencies will pursue those opportunities, but with discipline. The Office of Management and Budget’s Memorandum M-24-10 directs agencies to advance AI innovation while managing risk, and it requires stronger governance for uses that could affect rights or safety. The National Institute of Standards and Technology’s AI Risk Management Framework and its generative AI profile give agencies a practical structure for testing, monitoring and governing those systems.

That combination should shape how government interprets the economics. The real question is not whether a model is cheaper than a contractor hour in the abstract. The real question is where AI can lower the cost of safe, reviewable work without weakening accountability, records management, privacy protections or public trust. In that sense, the most valuable federal uses may be less glamorous than public debate suggests. They are the internal workflows that free analysts, attorneys, acquisition staff and program managers to spend more time on judgment and less time on mechanical first passes.

The bigger shift is inside the workforce, not the invoice

The Stevens paper tracks external spending, which makes it useful but incomplete for public institutions. Agencies often feel the impact of AI before it appears as a neat line-item swap. The change shows up in shorter clearance cycles, fewer contractor hours for routine deliverables, delayed backfills and higher throughput from the same team.

That broader pattern is visible in government’s own reporting. According to The Government Accountability Office’s July 2025 review of federal agency practices, the number of reported AI use cases across the selected agencies it examined rose from 571 in 2023 to 1,110 in 2024, while generative AI use cases increased from 32 to 282. The Federal AI Use Case Inventory shows the same direction of travel: Agencies are moving beyond experimentation and into operational use.

That matters for workforce planning because the first-order effect of generative AI is often task compression, not immediate headcount reduction. An analyst who can turn a week of document review into an afternoon changes the economics of the unit even if no position disappears. Over time, however, those productivity gains can reshape hiring patterns and career ladders. A recent Stanford Digital Economy Lab working paper found a 16% relative employment decline for workers ages 22 to 25 in highly exposed occupations after the rise of generative AI. That is private-sector evidence, but the lesson travels: Entry-level work built around drafting and routine synthesis is becoming less secure unless it is paired with domain knowledge, evaluation and accountability.

For agencies, this creates a management challenge. The federal workforce still needs junior talent, but those roles must be redesigned. Early-career employees should spend less time producing raw first drafts and more time learning how to evaluate model outputs, protect sensitive data, apply policy context and exercise professional judgment.

Why governance is now a productivity strategy

Government professionals often hear governance framed as a brake on adoption. In practice, it is becoming the opposite. Clear rules are what let agencies scale useful AI instead of trapping it in pilots.

That is especially true because generative AI is increasingly behaving like a general-purpose technology. The OECD’s recent work on generative AI as a potential general-purpose technology argues that its significance lies in pervasiveness, ongoing improvement, and its ability to spawn complementary innovations. The OECD’s latest cross-country data on firm adoption shows that organizational use is rising quickly, especially in large enterprises. For government, that means the pressure will not fade. Mission teams, vendors and oversight bodies will all expect more capable use of these tools.

The agencies that benefit most will be the ones that treat governance as operating infrastructure. That means approved use cases, human review thresholds, logging and monitoring, model evaluation, procurement language for data handling, and training that teaches staff when to trust a tool and when to stop. CIO Council guidance on federal AI inventories and safeguards points in that direction by tying transparency and safeguards to actual deployment.

The point is simple: Cheap generation without governance creates rework, risk and political exposure. Cheap generation with governance can create faster service delivery, better internal operations and more capacity for mission work.

The most important number in today’s AI debate may not be model accuracy or headline valuation. It may be the three cents of AI spend that now appears where a dollar of contracted knowledge work used to sit. For government professionals, that figure is not a mandate to automate recklessly. It is a signal that the economics of routine cognitive work have changed, and that agencies need to respond with equal parts ambition and control. The winners will not be the agencies that buy the most AI. They will be the ones that redesign work, modernize procurement, and build governance strong enough to turn lower-cost output into higher-confidence public service.

—

Dr. Gleb Tsipursky serves as the CEO of the AI consultancy Disaster Avoidance Experts.

Copyright © 2026 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.