The federal government needs thousands of AI-capable staff it cannot hire. The mandates are real, the compensation ceilings are statutory, the pipeline wasn't built, and the programs meant to fix it don't match the scale of the problem. Your buyer lives inside this math every day, but they almost certainly haven't assembled the numbers into a single picture. That's what you're bringing to the meeting.
What the mandates actually imply
OMB Memorandum M-25-21 requires 80+ agencies to designate Chief AI Officers, stand up governance boards, publish compliance plans, and run risk assessments on every high-impact AI use case. Each of those requirements implies headcount. GSA's compliance plan, one of the more detailed public examples, describes a CAIO chairing an AI Governance Board, a co-chairing CDO, and a separate AI Oversight Committee reviewing every AI request for risk, privacy, and security. That's a minimum of three to five dedicated positions per agency for a functioning governance structure, before you count implementation staff.
By rough arithmetic, 80+ agencies times three to five governance-specific positions puts you at 240 to 400 dedicated roles needed governmentwide. No official OPM or OMB estimate exists for this specific number. It's a conservative floor based on minimum viable staffing, and it doesn't include the people who actually build, test, or operate the AI systems being governed.
So much for demand. Here's supply.
The compensation ceiling
A GS-15 Step 10 in the DC locality pays $197,200. Not a negotiable figure. It's a statutory cap under 5 U.S.C. §5304(g)(1), which prevents locality-adjusted GS pay from exceeding Executive Schedule Level IV. The uncapped formula would produce $220,002 for that position. Steps 7 through 10 all pay the same $197,200, which means a GS-15 in DC hits their compensation ceiling roughly mid-career and stays there.
The mid-career grades are where the structural gap opens widest. A GS-14 in DC tops out around $187,000. A GS-13 ranges from roughly $122,000 to $158,000. OPM Director Scott Kupor has stated that Tech Force recruits will mostly enter at GS-13 or GS-14, with salaries between $130,000 and $195,000.
Compare that to the private market:
| Federal (GS, DC locality) | Private Sector (DC metro) | |
|---|---|---|
| Mid-career AI/ML engineer | GS-13: $122K–$158K | 75th pctl: $214K (Glassdoor) |
| Senior AI/ML engineer | GS-15 Step 10: $197K (statutory cap) | 90th pctl: $259K (Glassdoor) |
| Senior, total comp (incl. equity) | $197K (no equity mechanism) | $280K–$400K (Kore1 placement data) |
| Median ML engineer, total comp | — | $245K–$265K (Levels.fyi) |
A source note on the private-sector numbers: Glassdoor and Levels.fyi are crowdsourced, self-reported, and skew toward people motivated to share compensation data. Kore1 reflects actual placements but publishes ranges that serve a recruiting function. The figures are directionally reliable, not precise. The structural point holds regardless of which source you trust most: the federal government cannot offer equity, and equity is where the real gap lives. A three-year AI engineer in the private DC market can earn what a GS-15 earns at the statutory maximum. Same job titles, completely different compensation architecture.
The pipeline compounds the problem
Brookings researcher Valerie Wirtschafter analyzed over 165,000 federal job postings pulled from the USAJobs API between January 2016 and March 2026, filtering to seven technical job series — six of which OPM had flagged as relevant to AI and AI-enabling roles. Of the 56,000+ postings that qualified as technical, fewer than 1,600 mentioned AI capabilities. Under 3%.
Sit with that for a second. For a decade, across the entire federal technical hiring apparatus, 97% of job postings didn't ask for AI skills. The government wasn't selecting for this capability because it wasn't planning for this capability. And then the mandates arrived.
The Brookings data also shows that roughly 25% of AI-specific listings appeared from 2024 onward, reflecting the hiring surge around EO 14110. The recent uptick is real, but it's building on a base that was nearly zero.
Two notes for the reader who may cite these figures. First, methodology: the 56,000 number represents the technically-filtered subset of 165,000 total postings. The keyword dictionary used to identify AI references was LLM-generated and then human-reviewed. The study doesn't address whether USAJobs re-postings were deduplicated. These are reasonable limitations for a think-tank policy analysis, not disqualifying ones. Know them before you put the number in a slide. Second, temporal context: federal hiring conditions shifted materially in 2025, and the 2016–2024 baseline in this data may not track forward. If you're citing these figures in Q3 or Q4 conversations, confirm whether the trend line still holds.
Program throughput vs. mandate demand
OPM is running three programs to address the gap. All three are genuine, and none of them operate at the scale the mandates require.
U.S. Tech Force: Annual cohorts of 1,000 fellows across AI, cybersecurity, data science, and software engineering. About 35,000 expressed interest; ~10,000 submitted full applications. Agencies are in final hiring stages for the first cohort. These are 1- to 2-year fellowships. The OPM memo does not address conversion to career positions. If fellows aren't converted, the program adds temporary capacity that walks out the door on schedule.
Data Science Fellows: 250 initial target, expandable to 1,000. Launched Spring 2026. No enrollment numbers published yet. The program targets data science broadly, not AI governance specifically.
Early Career Talent Network: Launched April 2026. Initial focus: finance, HR, engineering, project management, and procurement. Not AI. Not security. Not identity. The first cohort targets 200 student interns. Kupor has acknowledged the structural problem directly:
"Massively under-indexed on early career talent" — by a factor of 3:1, with only about 7% of the federal workforce under 30.
Tech Force delivers ~1,000/year across four disciplines, temporarily. Data Science Fellows targets 250. ECTN isn't aimed at AI roles. Total AI-relevant throughput: a few hundred per year, split across 80+ agencies.
What this means for your Tuesday call
Your buyer feels this constraint daily. They have a CAIO mandate, a governance board requirement, compliance deadlines, and a hiring pipeline that delivers a fraction of what they need on a timeline that doesn't match their reporting obligations. What they probably haven't done is put the compensation math, the Brookings pipeline data, and the program throughput numbers next to each other. You're giving them arithmetic for a frustration they've only experienced as understaffing.
The discovery questions that surface this need to be specific enough that the buyer can't deflect with "we're working on it."
"How many people on your team are currently responsible for reviewing AI use cases for access and authorization compliance?" Gets at headcount directly. The answer is almost always a small number doing it as a collateral duty.
"When a new AI tool gets provisioned, what does the access review process look like today, and who owns it?" You're looking for manual processes, ad hoc ownership, and the absence of a dedicated workflow. If the answer involves spreadsheets or email chains, you've found the pain.
"If your CAIO mandate requires annual use case inventories with risk assessments, how are you tracking which AI systems have access to which data stores, and who reviews that mapping?" Connects the governance mandate directly to identity infrastructure. The buyer who can't answer this cleanly is the buyer who needs automation.
"Given the compensation constraints on GS hiring, are you planning to staff your way to AI governance capacity, or are you looking at infrastructure that reduces the per-review human burden?" Use this with a buyer who's already acknowledged the staffing problem. It opens the door to an architecture conversation.
All four questions point to the same structural reality. Your buyer cannot hire fast enough, at competitive rates, through programs that are temporary and undersized, to meet mandates that are permanent and expanding. Every AI use case that comes online creates a new access relationship that someone has to govern. Automated provisioning and deprovisioning, lifecycle management that triggers on policy rather than on a person remembering to file a ticket, access reviews that run continuously instead of quarterly when someone finds the time. These are what turn a three-person governance team into one that can cover the workload of fifteen people the agency will never be authorized to hire.
That's the math. Bring it to the meeting.
Things to follow up on...
-
85% missing risk documentation: Brookings found that more than 85% of high-impact deployed federal AI use cases in 2025 lack required risk mitigation information, which means the governance gap this piece describes is already showing up in the compliance data your buyer has to report.
-
13,000 IT managers lost: The 2210 IT management job series saw one of the highest separation rates in government last year, with roughly 13,000 departures governmentwide including 4,300 at DoD alone, compounding the AI governance staffing problem with a loss of the people who understood legacy access architectures.
-
DHS continuous authorization signal: DHS committed in its published AI strategy to shifting toward a continuous authorization model for AI systems rather than point-in-time ATOs, which is a concrete policy signal that maps directly to real-time access governance and gives sellers a specific reference point in DHS account conversations.
-
M-26-10 is FITARA for AI: OMB's March 2026 memo requires CIOs to submit contract data on all IT purchases including AI to OMB starting May 2026, a shadow-AI enforcement mechanism that will force visibility into AI spend your buyer's CIO may not currently have.