EXPOSED
The daily work — pulling case law, summarizing records, cite-checking, and producing first-draft bench memos and opinions — is exactly the text-in/text-out labor that current models handle at usable quality, and courts are already piloting AI research tools. What survives is the confidential apprenticeship with a specific judge: reading a hot bench, flagging the weak link in a party's theory, and being a trusted second mind whose judgment the judge tests ideas against. The role is also insulated by institutional tradition and fixed judiciary headcount rather than by market economics, so the count holds even as the task mix hollows out toward review-and-challenge.
Part 2020 shock, part continued decline in the years since.
Median pay $54,010 → $64,920 -3.8% 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
+2.5% 14,500 → 14,900 on the projections basis
Exposed, but growing
AI can already do a lot of these tasks, and the BLS still expects +2.5% 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.
~1,000 openings a year on average, including replacing people who leave.
Law ClerkLegal ClerkLaw AssociateChancery ClerkDistrict ClerkJudicial ClerkLaw ResearcherTerm Law ClerkCareer Law ClerkPro Se Law ClerkFederal Law ClerkAttorney Law ClerkJudicial AssistantJudicial Law ClerkAppellate Law ClerkLaw Firm ConsultantFamily Law AssociateState Appellate ClerkFederal District ClerkFederal Appellate ClerkDistrict Court Law ClerkCareer Judicial Law ClerkCourt of Appeals Law ClerkFederal District Law Clerk
Holding it up: trust premium . Weakest point: embodiment .
Mixed — a routine tier and a judgment tier Drafting bench memos, Westlaw/Lexis research, cite-checking under Bluebook, and summarizing the record are precisely what a long-context model does at draft quality, which pulls this down toward 6; it sits at 8 rather than lower because oral-argument prep, sifting a 3,000-page administrative record for the fact the parties buried, and pushing back verbally when the judge floats a theory in chambers still require you in the room.
Fully desk- and screen-based The job is the screen, the docket, and the printed brief on your desk — sitting in the courtroom during argument and walking the opinion draft down the hall to the judge is the entire physical footprint, which is why it registers a 2 instead of 0.
No licence, no signature requirement Nothing you write carries legal force until the judge signs it; the JD and bar admission are hiring conventions, not a statutory gate — clerkships routinely hire pre-bar-results, and if a clerk misses a controlling case the sanction lands on the court's reputation and the judge, not on your license, so the 4 reflects credential preference with no personal exposure.
Meaningful discretion You decide which precedent is genuinely controlling versus distinguishable and how to frame an unsettled question, and on a busy district docket you effectively triage what merits the judge's attention — real discretion, hence 10 — but every call is reviewed and can be overridden before it becomes an order, so you never own the outcome the way an Article III judge does.
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 (8/20 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 (4/20) is whether the law requires a licensed human to sign. Trust premium (11/20) is whether buyers specifically pay for a person. Judgment and accountability (10/20) 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 25 of this occupation's 35 points (71%).
Embodiment (2/20) 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.
Lawyers SAFE
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 48/100, still EXPOSED.
A judiciary-wide rule making a named human — the clerk or judge — certify that every citation and quotation in a filed opinion was independently verified against the source, extending the AI-certification standing orders already issued by Judge Brantley Starr (N.D. Tex.) and adopted piecemeal in dozens of district courts, plus the Administrative Office's July 2025 interim AI guidance requiring human review of AI-assisted judicial work product. If the certification names the clerk personally and is enforceable by discipline, the drafting role acquires an attestation that cannot be delegated to a model.
Genuine two-tier structure: if commercial tools absorb pull-the-cases, summarize-the-record and cite-check, the residual day is adversarial review of machine drafts — finding the fabricated quote, the overruled holding cited as good law, the sub silentio conflict with circuit precedent — which is harder than producing the draft. Watch for chambers reorganizing clerk workflow around verification and bench-memo challenge rather than first drafts.
Chambers or Judicial Conference policy barring sealed records, grand jury material, presentence reports and draft opinions from any model not running inside the judiciary's own environment — the confidentiality argument already animating state court AI task forces (e.g. Illinois Supreme Court policy effective Jan 2025, Delaware and Texas judiciary interim policies). If draft opinions legally cannot leave chambers, the trusted-insider second mind is the only reader available.
Formalizing the clerk's dissent function: a chambers or circuit practice requiring a written clerk recommendation on record that is separate from the judge's and retained, so the call under ambiguity is attributed. Some appellate staff attorney offices already produce signed recommendation memos on screening panels; extending named-recommendation practice to elbow clerks would make the judgment call owned rather than invisible.
The limit. Headcount is set by fixed judiciary appropriations and one-to-two-clerks-per-judge tradition, not by output demand, so these levers protect the content of the job more than they protect the count — and conversely, the count could hold at 13,290 while the work becomes machine-review even if none of these levers fire. No lever raises embodiment.
| New York-Newark-Jersey City, NY-NJ | 560 | $58,890 -9% |
| Philadelphia-Camden-Wilmington, PA-NJ-DE-MD | 460 | $58,890 -9% |
| Atlanta-Sandy Springs-Roswell, GA | 390 | $61,840 -5% |
| Boston-Cambridge-Newton, MA-NH | 380 | $143,350 +121% |
| Miami-Fort Lauderdale-West Palm Beach, FL | 340 | $55,620 -14% |
| Riverside-San Bernardino-Ontario, CA | 340 | $80,420 +24% |
| Seattle-Tacoma-Bellevue, WA | 320 | $80,240 +24% |
| Salt Lake City-Murray, UT | 300 | $56,080 -14% |
| Boston-Cambridge-Newton, MA-NH | 380 | $143,350 +121% |
| Springfield, MA | 40 | $143,350 +121% |
| Albany-Schenectady-Troy, NY | 80 | $136,950 +111% |
We have no reported case of a named organisation automating this occupation. Not one deployment, not one announcement.
That is worth saying out loud next to a score of 35. The verdict above is about what the work exposes — what current AI could do to these tasks. It is not a claim that anyone has done it. For this occupation those two things have come apart completely: the capability argument is on this page, and the evidence column is empty.
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.