NIH Shrinks the Workforce, Grows the AI: Follow the $1 Bait Hook

I smelled it before I finished the first paragraph. That familiar federal cocktail: a budget axe swinging in one hand, and “innovation” cologne sprayed with the other. The kind of situation where Washington unplugs people and then acts shocked when a shiny tool shows up pretending to be a miracle.

NIH: more AI use cases, fewer employees

According to federal inventory reporting and agency talk around NIH’s AI adoption, the trend line is clear: staffing shrinks while AI pilots grow. NIH’s headcount fell to roughly 17,000 employees in early 2026, down by more than 4,000 from just over a year earlier. At the same time, NIH’s reported AI use cases climbed to 124 in fiscal 2025, up from 82 in 2024, based on the HHS AI use-case inventory published under federal requirements.

Brick’s F-150 math: tools are fine, replacing the crew is not

I am not here to boo a calculator. I like tools. AI can absolutely be a tool. But when an agency that deals in life-and-death science loses thousands of workers and then leans harder on pilots, you do not get “efficiency.” You get shortcuts, burnout, and dashboards screaming “ALL GOOD” while the oil light flashes.

NIH officials and speakers have described AI work spanning:

• Administrative tasks (like analyzing grant portfolios)
• Research support and lab work
• Clinical assistance

A lot of it is still pilot or pre-deployment, meaning it is revving in the parking lot, not hauling a trailer across the country. And NIH folks have been blunt that scaling is the hard part, where messy data, foggy rules, and real accountability come due.

The vendor swamp and the “$1 deal” worry

Now for the villain: procurement gravity. NIH, like other agencies, has leaned on bundled buying efforts through GSA, including OneGov, launched in April 2025 to treat the federal government like one customer. Sounds clean in theory. In practice, it can become the classic trap: cheap up front, expensive forever.

One NIH technology leader raised concern about the “drug dealer model” of $1 deals that later sunset. Translation: free samples today, renewal shock tomorrow, after your workflows and training are already chained to the platform.

Small language models, big leverage

NIH speakers have discussed building domain-specific small language models trained on large NIH datasets (including Alzheimer’s data) so researchers can ask questions within a tight, controlled domain. That direction is promising. Small and auditable beats giant black-box oracle.

NIH is also running a generative AI community of practice with roughly 2,000 people, pushing training and careful use (including human-in-the-loop and data protection). Good guardrail talk. But guardrails take staff, time, and spine, especially across NIH’s 27 institutes and centers.

If you drain the workforce and replace it with pilots, you are not modernizing. You are outsourcing responsibility and praying the discount never ends.