Why this work stays close to people
Teaching someone to cross a busy intersection with a long cane is not an information problem. It is a trust problem, a body-mechanics problem and a street-by-street judgment call. The specialist walks the route with the client, reads traffic sound, watches posture and steps in when a curb cut is missing. Software can describe a scene. It cannot take responsibility for a person’s safety at the corner.
The same holds for in-home work. A vision rehabilitation therapist walks through a kitchen, finds the three things that cause falls or burns, and rebuilds the daily routine around what the person can still see. Much of the teaching is physical: hand-under-hand guidance for braille, positioning a magnifier at the right distance, practicing a stove task until it feels ordinary. That is why the share of task time our method leaves to a person, 73%, sits where it does.
There is an emotional layer too. Many clients arrive soon after a diagnosis, angry or frightened. Pacing a lesson around grief is a skill no model has been tested on in this setting.
What AI does, assists with, and leaves alone
Some tasks already run well without a person in the loop. Converting print materials into large print, braille-ready files or audio is routine automation. So is a first pass at session notes, progress summaries and the paperwork that follows an assessment. The share our method marks as work AI can do on its own is 0%. Our coverage method explains how that time is measured.
A larger block of work is assisted rather than handed over. Planning a lesson sequence, tracking progress across weeks, and matching a client to a magnifier, screen reader setting or wearable can all start with a tool and finish with the specialist. Scene-description apps and smart glasses add a channel of information during travel practice; the instructor still sets the route and the goal. That assisted share is 27%.
What is left is the core: route training in live traffic, functional vision assessment in the client’s own home, braille and daily-living instruction, and the judgment of when someone is ready to travel alone. Those tasks carry physical risk and legal responsibility, which is why they do not move first.
Good to know: cheaper assistive technology usually expands who gets taught, because someone has to teach the device.
What the evidence actually shows
Not much, yet. The evidence grade on this page is D, and the lowest grade means the same thing everywhere on this site: no one has run a direct, published test of an AI system against qualified practitioners doing this job. Because of that, we publish no parity number here. Claiming one would be guessing.
What would settle it is specific. A controlled trial comparing independent-travel outcomes for clients trained with an AI-guided navigation system against clients trained by a certified orientation and mobility specialist. Or a study measuring whether device-matching by software produces the same reading speed and task completion as a certified low vision therapist’s recommendation, six months out. Until work like that exists, the honest answer is that the field is untested rather than proven either way. The scoring method keeps the ungraded case separate from the measured one on purpose.
The outside labor data is clearer. BLS reports about 162,450 people employed in this occupational group and median pay of $100,330 (BLS, 2025), with projected employment growth of about 14.8% between 2025 and 2035. An aging population with more age-related eye disease is the main reason demand holds up.
When the timing could shift
Most likely after 2042 (8 in 10 of our scenarios). Our replacement-year method sets out what that window does and does not mean.
Two things could pull it earlier. First, cost: running an AI tool for a year is a fraction of what it costs to employ a person, and the cost panel above shows that gap. Second, wearables. If scene description, obstacle warning and route guidance become reliable enough that clients need fewer supervised hours, caseloads stretch further and fewer new instructors get hired. That is task erosion, and it usually shows up in entry-level openings first.
Two things hold it back. The robotics requirement is the big one: the hardware tier for the physical part of this job is rated a dexterous humanoid, which does not exist as a deployable product. And liability does not transfer. A funding body, a school district or a VA program signs off on a certified professional, not a model. Certification through ACVREP, state contracts and third-party billing all assume a named human is accountable for the outcome.
How to stay needed
Lean into the parts of the job that stay with a person. Live route training and street crossings in unfamiliar environments. Home and workplace assessments, where the answer depends on lighting, clutter and habits you can only see in place. Braille and tactile instruction, where feedback travels through the hands.
Two skills raise your value quickly. One is assistive technology fluency: knowing current screen readers, OCR apps, electronic magnifiers and AI wearables well enough to teach them, troubleshoot them and tell a client when a device is the wrong fit. The other is case coordination with ophthalmology, employers, schools and funders, which is relationship work before it is documentation.
If you are weighing adjacent paths, the closest work sits with occupational therapists, rehabilitation counselors and orthoptists. You can put any two of them side by side on the job comparison tool, or look at the wider diagnosing and treating practitioners family to see how the task splits differ.
Our headline figure for this job, 78 out of 100 (higher is safer), reflects a job where the hands-on majority of the work has nowhere cheap to go. Hiring pressure is the thing to watch, not disappearance. For context on demand across care roles, see the healthcare sector page and our list of in-demand jobs.