Optometry AI comes in 3 waves, and only the first is ready to buy

Three waves, three timelines, and only one of them is something an independent optometry practice can buy today. That was the framing from a panel at Academy 2026 Anaheim, reported by Vision Monday on 5 October 2026: operational AI is current, clinical AI is one to three years out, and interventional AI is five or more years away. If you direct a practice with several optometrists, a governance duty and a stack of vendor emails, that sequence is the shortest ranked list you are likely to get this quarter.

The panel, hosted by Jobson Optical Group, included Scot Morris, OD, Howard Purcell, OD, president and CEO of New England College of Optometry, Chris Wroten, OD, and Carl Spear, OD, of EssilorLuxottica. Spear's employer sells lenses and owns retail optical brands, so treat the commercial vantage point as part of the picture. The structure they offered is still useful, because it tells you which purchase decisions carry clinical risk and which carry only a wasted subscription.

Wave one is operational, and it is the only wave with products you can sign for today

The panel placed administrative automation, ambient documentation, pretesting and coding or billing support in the current wave. These tools touch the front desk, the chart and the claim. None of them makes a diagnosis. A failure costs you time, money or a compliance headache, not a missed pathology.

Documentation is where the newest launches sit. RevolutionEHR announced an AI Scribe for the exam room on 27 April 2026, described in its own release as hands-free voice capture that fills structured encounter fields and asks the provider to approve changes before they are final. The release says the feature comes at no additional cost, though it does not say whether that applies to every plan. It gives no time-saving figure, no accuracy data, and no statement on HIPAA or a business associate agreement for the scribe (the only HIPAA reference on the page relates to a different product). All of that is vendor material, so none of it is evidence yet.

That gap is the useful finding. A free scribe inside your existing EHR is the easiest pilot you will ever be offered, and it is also the easiest one to start without asking where the audio is processed, whether it is retained, and what patients are told. Before any exam-room recording begins, get the vendor to put the answers in writing, on the exact plan you use. We covered the same plan-level trap for therapy scribes in our check on free AI scribe tiers and BAAs, and the logic transfers directly to an optometry chair.

So what for you: if you buy anything this year, buy from wave one, and make "signed BAA for this plan, recording consent script, provider review before sign-off" the three conditions on every pilot.

Wave two is decision support, and the risk shifts from your budget to your judgment

The panel's second wave covers predictive risk modeling, biomarker analysis and clinical decision support, with confidence scores for glaucoma given as the example. The estimated arrival is one to three years. That does not mean nothing exists now. It means the category is still maturing, and most of what exists is assistive: it prompts a clinician, who makes the call.

The distinction between assistive and autonomous matters more than any feature list. In our August 2026 analysis of optometry's one cleared autonomous AI pathway, diabetic retinopathy screening stood alone as the condition with an FDA-cleared autonomous system and a dedicated billing code, while glaucoma and macular degeneration tools were decision support only. Nothing in this week's reporting changes that picture. A confidence score on a glaucoma risk model is a second opinion you chose to read, and the liability for the decision stays with the optometrist who signed the chart.

That is why clinical governance should come before clinical purchasing. If two optometrists in your group would react differently to the same AI flag, one escalating and one ignoring it, you have a protocol gap that no software closes. Write down who reviews a flag, what happens when you disagree with it, and how you record that you did. A one-page rule is enough.

So what for you: do not buy wave-two software until your clinicians have agreed what they do when the software and their own reading disagree.

Wave three is a forecast, not a product category

Autonomous diagnostic agents and AI-guided interventional procedures were placed five or more years out. Treat any vendor pitching them as available today with the skepticism you would apply to a miracle supplement. The panel also floated a longer-term shift in which history-taking and some diagnostic steps move outside the exam room, leaving the optometrist closer to a coordinator of the patient's care than a gatekeeper.

You can already see a crude version of that idea outside the practice. Eyebot, a Boston company, places self-serve kiosks that run an automated vision test in about 90 seconds, according to the company, measuring refraction, lensometry and visual acuity with a remote licensed provider reviewing results before a prescription is finalized. Reporting by Eyes On Eyecare on 3 February 2026 said they were in malls, universities and retail stores in four states, with a December 2025 pilot in Pennsylvania Walmart Vision Centers and Sam's Club Optical Centers, a $25 administrative fee, and an age range of 18 to 64. The kiosks cannot measure intraocular pressure. The company cites two IRB-approved studies with nearly 800 participants for its autorefractor, which is its own evidence and has not been independently confirmed here.

The American Optometric Association has asked the FTC and FDA to look at these kiosks and revised its telemedicine policy to say direct-to-patient technology should not replace a full eye exam, per the same report. The Pennsylvania Optometric Association issued a patient safety alert in January 2026. You do not need to share the profession's alarm to see the commercial point: a patient who has done a 90-second refraction at a retail store arrives with a number and an expectation.

So what for you: prepare your front desk and your clinicians for patients who arrive with kiosk results, and decide in advance how you explain what the kiosk did not test.

Why the stakes are higher for independents than the panel suggests

The ownership data adds pressure. According to the American Optometric Association's workforce figures, reported by Review of Optometry in January 2025, private practice's share of the optometric workforce fell from 51% to 41% between 2017 and 2024, while the private-equity office share rose from 3% to 10.9%. Larger groups can fund a vendor evaluation and a governance lead. Most independents cannot, which makes a short ranked sequence worth more than a long feature comparison.

The access argument the panel raised cuts the other way. Spear noted that many counties in Texas and Florida have no eye doctor, and Morris said he is the only OD in six counties. In those places, an operational AI that returns hours to a stretched clinician is not a convenience. It is capacity the community does not otherwise have. Setting the first goal as "more patients seen per week without longer days" is a better test of a wave-one purchase than any demo.

So what for you: measure a wave-one pilot in clinician hours returned and patients seen, not in feature count.

A sequence a practice director can adopt this month

The evidence points toward a staged approach, and frankly, if your group has no written AI policy today, the math favors writing that before any purchase. Here is a sequence that fits the three waves.

Start with the documentation and admin layer you already pay for. Ask your EHR and practice management vendors what AI features are on your current plan, get BAA and recording-consent answers in writing, and pilot with one clinician for 30 days against a baseline you measured beforehand. Next, adopt a one-page rule for any clinical AI output: who reviews it, what you do if you disagree, and where that is recorded. Last, if diabetic retinopathy screening fits your patient mix, evaluate the one cleared autonomous pathway on its own merits, using the checks in our earlier piece. Leave everything else on a watch list, reviewed twice a year.

For a scored, independent read on which wave your practice sits in and what to do first, the AI Opportunity and Growth Assessment covers exactly this kind of sequencing, and you can start with a free 20-minute discovery call.

So what for you: you are not behind. Sequence the decision, and a cautious practice can set the pace instead of reacting to the next kiosk or vendor email.

Frequently asked questions

Is there AI an optometry practice can buy today? Yes, in the operational wave: documentation, scheduling, pretesting support and billing support. Clinical decision support is maturing, and autonomous diagnosis has a cleared pathway only for diabetic retinopathy screening.

Is an AI scribe in my optometry EHR HIPAA compliant? HHS does not certify products, so the answer depends on your contract. Ask for a signed business associate agreement that names the exact plan, and check recording-consent rules in your state. The RevolutionEHR release reviewed for this article made no BAA statement for its scribe.

Do patient eye-test kiosks replace an eye exam? Per the American Optometric Association, direct-to-patient technology should not replace a full exam. The Eyebot kiosk described in February 2026 reporting cannot measure intraocular pressure, which is used to assess glaucoma risk.

Should I wait for clinical AI before starting anything? No. Operational pilots carry the lowest clinical risk and build the governance habits you will need for wave two. Wait on clinical purchases until your clinicians agree how they handle a disagreement with the software.

Sources: Vision Monday, 5 October 2026 (Academy 2026 Anaheim panel); RevolutionEHR / PracticeTek release, 27 April 2026 (vendor material); Eyes On Eyecare, 3 February 2026; Review of Optometry citing American Optometric Association data, 15 January 2025. This article is general information, not clinical, legal or regulatory advice. Confirm product claims and your data protection duties with the vendor and your counsel before acting.

See also: our CARE Framework comparison of AI scheduling tools built for optometry.

Related: Optometry has exactly one cleared autonomous AI test · Free AI scribe tiers rarely include a BAA · Optometry lost 10 points to consolidation