SAFE
The diagnostic image-reading layer — fundus photos for diabetic retinopathy, OCT for macular degeneration, visual field interpretation — is where AI is already FDA-cleared and genuinely good, and that will compress screening volume and referral triage. But the revenue core is microsurgery: phacoemulsification, intravitreal injections, laser capsulotomy, retinal detachment repair, performed on a moving, anesthetized human eye with sub-millimeter tolerances no current robot handles autonomously. State licensure plus personal malpractice exposure means a physician signs every surgical plan and every off-label injection decision.
Most of this decline happened after 2021 — it is not the pandemic dip.
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.
BLS projection, 2024–2034
+4.3%
Percentage only. The projection counts a different population from the 8,950 above — it includes self-employed workers, which for this occupation is most of them, so the two headcounts are not comparable.
Hard to automate, and growing
The work resists current AI and the BLS projects +4.3% more of these jobs by 2034. Note that safe does not mean well paid — several of the fastest-growing resistant occupations are among the lowest paid on the register.
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.
~300 openings a year on average, including replacing people who leave.
PhysicianOphthalmologistCornea SpecialistRetina SpecialistOphthalmic SurgeonRefractive SurgeonGlaucoma SpecialistMedical Doctor (MD)Neuro-OphthalmologistOculoplastic SpecialistOphthalmology PhysicianClinical OphthalmologistSurgical OphthalmologistOphthalmologist SpecialistVitreoretinal Disease PhysicianCornea and External Disease Physician
Holding it up: liability shield . Weakest point: task resistance .
Tasks largely resist digitisation Cataract extraction, anti-VEGF injections into the vitreous, YAG capsulotomy, trabeculectomy and scleral buckling are the daily schedule for most comprehensive ophthalmologists — and while the slit-lamp exam grading and OCT/fundus interpretation feeding those decisions is genuinely being automated, that displaced reading layer is a real fraction of clinic time, which is why this sits at 14 and not 18 like a purely operative specialty.
Hands-on in uncontrolled environments You work with a hand on a joystick-mounted phaco handpiece inside a 12mm globe, tilt patients under an operating microscope, perform scleral depression and indirect ophthalmoscopy on uncooperative or photophobic patients, and manage intraoperative bleeding or capsule rupture in real time; it stops short of 20 only because a meaningful share of the week is clinic-based refraction, imaging review, and dictation at a desk.
Licensed human required and personally liable State medical licensure, ABO board certification, ASC and hospital credentialing, and DEA registration for controlled post-op analgesia all gate the work, and the operative note plus the informed-consent discussion for endophthalmitis and retinal-detachment risk carry your name — a wrong-IOL-power implant or a missed giant cell arteritis is your malpractice claim, not the technician's who took the biometry.
Exists to be accountable for ambiguous calls Deciding whether a macula-off detachment goes to the OR tonight, whether to continue aflibercept in a non-responder or switch off-label, when to stop tapering steroids in uveitis with rising IOP, and whether a monocular patient's cataract density justifies surgical risk at all — these are calls made on incomplete evidence with irreversible outcomes, though many high-volume days follow well-established post-op and glaucoma-monitoring protocols.
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 (14/20 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 (19/20) is whether the law requires a licensed human to sign. Trust premium (16/20) is whether buyers specifically pay for a person. Judgment and accountability (17/20) 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 52 of this occupation's 83 points (63%).
Embodiment (17/20) 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.
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 89/100, still SAFE.
Task-mix shift: as FDA-cleared autonomous screening (IDx-DR/LumineticsCore, EyeArt) absorbs routine diabetic retinopathy grading and normal-OCT triage, the residual clinic day concentrates on ambiguous cases — atypical uveitis, indeterminate choroidal lesions, surgical candidacy in comorbid eyes, complication management — which current models handle poorly. The genuinely two-tier structure here (screen vs. decide-and-operate) makes this a real rise, not a reframe.
If malpractice carriers (OMIC, the ophthalmology-specific mutual) condition coverage on documented physician override review of AI screening outputs — an underwriting condition already appearing in radiology AI policies — the role formally owns the false-negative call.
If CMS finalizes reimbursement rules for autonomous AI retinal screening that require an attributable physician of record for any AI-flagged referral, and/or state medical boards adopt language (as some have for teleradiology) that AI output is 'preliminary' until a licensed ophthalmologist attests, the sign-off requirement becomes explicit rather than customary.
The limit. Already 83/100; realistic headroom is a few points. The binding constraint is surgical embodiment and licensure, both saturated — score movement is mostly noise unless autonomous intraocular robotics arrives, which would move it down, not up.
| Chicago-Naperville-Elgin, IL-IN | 510 | $284,630 -5% |
| Boston-Cambridge-Newton, MA-NH | 330 | $325,200 +8% |
| Miami-Fort Lauderdale-West Palm Beach, FL | 300 | $239,200 -20% |
| Dallas-Fort Worth-Arlington, TX | 210 | $359,980 +20% |
| Providence-Warwick, RI-MA | 200 | $399,020 +33% |
| Houston-Pasadena-The Woodlands, TX | 180 | $320,450 +7% |
| Baltimore-Columbia-Towson, MD | 130 | $210,490 -30% |
| Nashville-Davidson--Murfreesboro--Franklin, TN | 120 | $232,300 -23% |
| San Jose-Sunnyvale-Santa Clara, CA | 60 | $700,850 +134% |
| Atlanta-Sandy Springs-Roswell, GA | 60 | $412,900 +38% |
| Providence-Warwick, RI-MA | 200 | $399,020 +33% |
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 83. 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.