Rapidly Scale U.S. Government AI Capability

In 2025, Washington made frontier AI available to every federal agency for $1 each, and agencies' reported uses of generative AI rose ninefold in a year. The people who can judge those tools are scarce, and the teams that once supplied them have come and gone. Congress should require agencies to measure results, share tools and contract terms, hire technologists durably, and put them on real public services.

The Problem

In August 2025, the General Services Administration secured ChatGPT Enterprise for any federal agency at $1 per agency,¹ and Claude for all three branches at the same price.² At 11 agencies with AI inventories, reported AI uses nearly doubled from 2023 to 2024, and generative AI uses rose from 32 to 282.³ The Department of Veterans Affairs has begun automating medical-imaging work that supports veterans' diagnoses.³

Buying the tool is the easy part. Someone must test whether it works, protect sensitive data, notice errors, and decide when to stop. That takes technologists, and government has few young ones: roughly 7% of federal employees are under 30, compared with about 22% of workers outside government.⁴ Washington can license a frontier AI system for $1; hiring one employee takes an average of 101 days.⁵ Three weaknesses make this dangerous:

1. Skills lag behind use. Six of the 12 agencies GAO studied reported trouble attracting and developing generative AI expertise, competing with private employers for the same people.³
2. The $1 era is ending. OpenAI's $1 agreement expires at the end of September 2026; its replacement, a discount on usage-based pricing, is expected by trade press to raise agencies' costs.⁶ Agencies must now judge what the tools are worth.
3. Tech teams come and go. GSA's 18F completed 455 projects for 34 agencies, then was eliminated as "non-essential" in March 2025.⁷ OPM's new Tech Force aims to hire about 1,000 technologists for two-year terms.⁸

Why legislation: Congress has already required agencies to inventory their AI uses⁹ and to train acquisition staff in AI,¹⁰ and in 2020 it created an AI Center of Excellence at GSA, whose authority lapsed under a five-year sunset in December 2025.¹¹ OMB has "strongly encouraged" agencies to prioritize hiring technical AI talent,¹² but encouragement creates no positions and pays no salaries. Programs created by one administration can be dissolved by the next, as 18F was; durable capacity needs statute and appropriations. A government that uses powerful technology on the public's behalf owes the public people who understand it.

The Solution

A four-step staircase: each step stands alone, and each step up adds people, money, and responsibility. It covers agencies' own use of AI across civilian and defense missions; executive oversight of AI and whistleblower protections are addressed separately.

Step 1 — Count what AI delivers. For each significant use, agencies publish service quality, error rates, cost, and access measures against a baseline, with protected annexes for sensitive programs. Inventories already list uses; this adds results. A failed pilot should yield reusable lessons and a decision to stop or redesign, without pressure to declare success.

Step 2 — Buy once, share widely. Give agencies approved computing environments, model-evaluation services, procurement support, and reusable contract terms that guarantee data and model portability, audit access, and exit plans. OMB already tells agencies to avoid "costly dependencies on a single vendor";¹³ shared terms and shared evaluators make that possible for agencies too small to negotiate alone.

Step 3 — Hire the builders and the checkers. Fund permanent technical positions, paid fellowships, and governed exchanges with industry and universities, using skills-based hiring and pay flexibilities. Staff inspectors general, enforcement offices, and evaluation units with technologists too, and train procurement and program leaders in AI's limits. Require conflict screening, disclosure of outside interests including retained company stock, and a plan to keep knowledge when fellows leave.

Step 4 — Put teams on real services. Choose a manageable set of projects, each with a baseline, a public-service outcome, a responsible official, and a stopping rule. Pair engineers with frontline staff and the people served, and keep human review and appeal for consequential decisions. Success is a veteran who gets the right answer, not a count of employees with chatbot access.

Where to start: Step 1 is the floor; it asks agencies to report what they should already know. Step 3 is the heart: tools are cheap, and people who can judge them are scarce.

Administration and enforcement: OMB coordinates a five-year program and submits a costed workforce and investment plan within 180 days; GSA and OPM supply shared tools and recruiting, and agencies own delivery. Appropriations follow the staffing plan, with annual inspector-general review and a GAO evaluation before renewal.

Risks and Mitigations

• Spending without results: Technology programs can fund activity instead of outcomes. Fund projects in stages, evaluate them independently, and keep authority to cancel; training completions and tool adoption will not count as success. Some projects will still fail.
• Revolving doors: Industry exchanges speed hiring but invite conflicts; Tech Force members may seek jobs with partner companies when their terms end.⁸ Require screening, recusal, disclosure of retained equity, and government ownership of essential work products. The pull of future employers remains.
• Faster mistakes: Capable teams can also deploy flawed systems faster. Keep privacy, civil-rights, records, security, and appeal obligations in force; real capability includes the judgment to reject a system.

Similar Bills

Fit measures similarity to this proposal's mechanisms: High = direct precedent; Partial = useful component with material differences; Related = adjacent approach.

Federal

State

No state program comparable to a federal technical corps was verified in this review; the state rows show reporting and investment components only.

What this adds: Existing law tells agencies to inventory AI and train buyers, and the statute that created GSA's AI center has lapsed. This proposal adds measured results, shared tools and contract terms, durable funding for technical staff in agencies and their watchdogs, and delivery teams held to public-service outcomes.