COOKED
The core loop — collecting application data, cross-checking income and residency documents against wage and asset databases, applying deterministic program rules, and generating notices of approval or denial — is exactly the rules-plus-documents work that AI and self-service portals already handle at scale; many states have already moved SNAP and Medicaid renewals to automated ex parte processing. What resists is the messy human residue: applicants with no documents, unstable housing, language and literacy barriers, suspected fraud, and appeals where a caseworker's read of a situation matters and due-process rules require a human decision-maker on record. The modal worker is a county or state caseworker in a benefits office, and that job shrinks toward a smaller tier of complex-case and hearings specialists.
Headcount grew steadily across the period.
Median pay $46,590 → $54,210 -6.9% in real terms
This line is counted by the Bureau of Labor Statistics — the one figure on this page that isn't a judgement of ours. Headcount moves on demand, offshoring, demographics and the business cycle, and automation is one term among several, often not the loudest.
So a falling line is not evidence that AI did it, and a rising one is not evidence that it won't. Both happen in this register: some occupations resist automation and shrink anyway, others are highly automatable and keep growing. The marked year is 2020.
BLS projection, 2024–2034
+1% 166,800 → 168,500 on the projections basis
Exposed, but growing
AI can already do a lot of these tasks, and the BLS still expects +1% more of these jobs by 2034. Demand for the output is growing faster than the work is being automated away — the mechanism BLS gives for software developers, and the combination people most often misread as an error.
Different clocks. The score is what current AI could do to this work today. The projection is how many of these jobs will exist in 2034. Everything between the two — how fast employers actually adopt, whether demand grows in the meantime — is why they can point opposite ways without either being wrong.
~14,000 openings a year on average, including replacing people who leave.
InterviewerCase ManagerIntake ClerkContact AgentGrant ManagerGrant OfficerContact OfficerTenant SelectorCase CoordinatorEnrollment AgentEligibility ClerkHousing CounselorIntake SpecialistWorkforce AdvisorEligibility WorkerHousing SpecialistProgram SpecialistSocial Case WorkerEnrollment ProviderPlacement SecretaryReferral SpecialistWelfare InterviewerEligibility ExaminerEnrollment Counselor
Holding it up: trust premium . Weakest point: embodiment .
Core tasks are already automatable Verifying pay stubs against state wage-match files, computing SNAP net income deductions from a fixed formula, and mailing the resulting adverse-action notice are steps already coded into MAGI-based Medicaid determinations and ex parte renewals — a 6 rather than a 2 because the no-paperwork applicant, the shared-household question, and the hearing prep still need a person to build the record.
Fully desk- and screen-based The day is a workstation, a document scanner, and a phone or lobby window; the only physical variance is a home visit or outreach table that a minority of these workers ever do, which is why this sits at 4 and not 0.
Certification preferred, not legally required No state licence gates the job — you are hired with a bachelor's or clerical experience and trained on the state eligibility manual — but 5 reflects that federal due-process rules make you the named decision-maker on the notice and hearings record, so removing the human entirely creates an appeals problem rather than a licensing one.
Meaningful discretion Most calls are the manual's calls — countable income, categorical eligibility, verification hierarchy — with real discretion showing up in good-cause exemptions, self-declaration acceptance when documents are unavailable, and whether a discrepancy gets referred to program integrity; that band of judgment is genuine but narrow and supervisor-reviewable, which is a 7 rather than a 12.
The verdict above describes this occupation as a whole. Almost nobody does the typical version of a job — tick what's actually in your week and see how your own mix sits.
Your task mix speaks to task resistance (6/20 here) — how much of the day's work current AI already does. That is the dimension the boxes above are about.
It cannot move the other three. Liability shield (5/20) is whether the law requires a licensed human to sign. Trust premium (8/20) is whether buyers specifically pay for a person. Judgment and accountability (7/20) is whether the role exists to own consequential calls. Those are facts about the occupation's standing, not about which tasks are in your week — a paralegal who does only trial exhibits still holds no licence. Together they are 20 of this occupation's 30 points (67%).
Embodiment (4/20) is also a property of the work rather than the worker, but we don't tag individual tasks as physical or not, so the picker can't tell you anything about it. That's a limit of this tool, not a claim.
Did we get the list right? Tell us what's missing — the tasks are written from the outside, and you're reading this from the inside.
Human Resources Managers EXPOSED
The moves above are yours to make. This is the other half: what would have to change in the world for the occupation itself to score higher. None of it is in any one person's gift, but it is where the floor actually comes from. Scores here are not a one-way ratchet. Only two of the five dimensions — task resistance and embodiment — track what machines can do. The other three track law, what buyers will pay for, and who is answerable, and those move in both directions, often in response to the same pressure AI creates. If every lever below landed, this occupation would score around 45/100 — EXPOSED.
Courts or CMS/FNS enforcement requiring that adverse actions (denials, terminations, ex parte renewal failures) rest on an identified human decision-maker who can be examined at a fair hearing — the pattern already litigated in Idaho's K.W. v. Armstrong Medicaid algorithm case and Arkansas's ARChoices ruling, plus Michigan's MiDAS unemployment fraud fallout. If a consent decree or state APA amendment bars automated adverse determinations without documented individual review, the residual role becomes the accountable signer on every denial.
Pure task-mix shift, no law needed: as ex parte renewals and portal intake absorb clean cases, the surviving headcount concentrates in fair-hearing preparation, fraud referrals, undocumented-income budgeting, homeless and mixed-status household cases, and disability/LTC asset assessments. The occupation is genuinely two-tiered, so the residual job scores higher even as the headcount falls sharply.
State legislation modeled on proposed 'automated decision systems' bills (e.g. Colorado SB 24-205-style, California ADMT rules, Washington's introduced ADS bills) that specifically prohibits final eligibility denial or benefit reduction by automated means and names a state employee of record for each determination. Unlike a license, this is a statutory human-in-the-loop mandate, which shields the same way.
Narrow route only: union contracts (AFSCME, SEIU local government locals) that bargain minimum caseworker-to-caseload ratios or 'no automation of eligibility determination' clauses, as some state contracts have done for call-center displacement. This buys headcount, not buyer preference — applicants do not choose their caseworker and cannot pay for a human.
The limit. Even with every lever, this tops out in the low 50s and — critically — the levers raise the score of the surviving job while doing nothing for headcount. The single largest force here is a public-payer employer under budget pressure that gains directly from automating its own staff away; there is no client willing to pay extra for a human, and states have shown they will pursue automated renewal even at the cost of erroneous terminations. Expect a much smaller, harder, more defensible occupation.
| Los Angeles-Long Beach-Anaheim, CA | 15,160 | $63,520 +17% |
| New York-Newark-Jersey City, NY-NJ | 6,920 | $52,980 -2% |
| Philadelphia-Camden-Wilmington, PA-NJ-DE-MD | 4,620 | $60,960 +12% |
| Atlanta-Sandy Springs-Roswell, GA | 4,380 | $47,530 -12% |
| Riverside-San Bernardino-Ontario, CA | 3,250 | $57,470 +6% |
| San Juan-Bayamon-Caguas, PR | 3,170 | $49,940 -8% |
| Fresno, CA | 3,150 | $59,300 +9% |
| San Francisco-Oakland-Fremont, CA | 2,850 | $82,960 +53% |
| San Jose-Sunnyvale-Santa Clara, CA | 1,240 | $97,900 +81% |
| San Francisco-Oakland-Fremont, CA | 2,850 | $82,960 +53% |
| Santa Rosa-Petaluma, CA | 240 | $80,330 +48% |
GovExec reports that Code for America, working with Anthropic, is piloting AI tools to support SNAP benefits eligibility work.
StateScoop reports Code for America has partnered with Anthropic to build AI tools intended for SNAP benefits caseworkers.
Has AI actually changed your work? One tap, anonymous, and the running tally is public. Nothing else is asked of you.
Rather than check back: get the digest and we'll tell you what changed — or watch a single occupation from its own page.