AI Scribes for Biomedical Technicians: What the VA Deal Leaves Out
On September 18, the Department of Veterans Affairs awarded an enterprise contract for ambient AI scribes, worth up to $775.72 million over five years, to Abridge and Knowtex. The scribes listen to the visit and draft the note, and the clinician reviews it.
One VA provider, quoted in Knowtex's announcement, said it was the first time in ten years they could "complete my notes and go home on time".
Now walk down to the biomed shop. In most hospitals, including plenty that are buying AI scribes right now, the people who keep the pumps, monitors and ventilators running still write their work orders at the end of the shift, from memory, at a keyboard.
Same hospital. Same documentation pain. Only one side got the fix.
What did the VA actually buy?
A scaled-up version of something it has already tested hard. The pilot started in October 2025 at 10 medical centers. By mid-September, ambient scribes had been used in more than 986,000 VA appointments, and fewer than 1% of veterans opted out. Knowtex says its clinicians alone have saved more than 450,000 hours of documentation work. That is the vendor's own number, so read it as a claim, but the direction is hard to argue with.
The new contract covers outpatient care: primary care, behavioral health, rehab and medical and surgical specialists, across more than 75 medical centers. No inpatient plan yet. And nothing for the people who keep the equipment running. That is not a criticism of the VA. Nobody else has done it either.
Why is the bench the same problem?
Because the pain has the same shape. The work happens first. The record gets written later, from memory, once the real job is done. We hear this from service leaders again and again, and it is why so many work orders close on a line like "good to go." Monty Gonzales has written in AAMI News about what thin CMMS records cost the field. The two-word note is not laziness. It is what you get when the only tool is a keyboard at the end of a long day.
And HTM can afford it even less than medicine can. The field has about 7,300 openings a year and roughly 400 new graduates, with about 40% of the workforce approaching retirement. Every hour a tech spends typing is an hour of wrench time a short-staffed shop does not get back. Every detail left out of the record is knowledge that walks out the door with the person who knew it.
Can you just hand a clinical AI scribe to a biomed?
No. The idea transfers. The tool does not. Four things are different at the bench.
1. There is no conversation to listen to. A clinical scribe listens to two people talking in a room. A technician is usually alone with a device, hands inside it, gloves on. There is no visit to overhear. The tech has to talk on purpose, a few words between steps: the error code when it shows up, the reading when the meter settles, the part when it goes in. An AI scribe for HTM has to work from that, not from a dialogue.
2. The note has a different reader and a different shape. A clinical note follows a format every clinician was trained on. A work order does not. It has to land in specific CMMS fields: failure code, parts, labor, resolution. And every shop has its own house format for how a good closing note reads. The shops we have talked to each have one, and no two match (more on that here). A scribe that writes a nice paragraph in the wrong shape just creates cleanup work.
3. Length cuts the other way. In medicine, the complaint is too much typing. At the bench, the complaint is also notes that are too thin. But here is a question worth asking before you buy any voice tool: if typing was the only thing keeping notes short, what happens when you take it away? Does the tech who used to write "good to go" now narrate the walk up the stairs? Shops that tried a minimum character count learned that padding is not detail. A replaced thermometer needs one line. A multi-step repair needs a real story. The right test is not length. It is whether someone who was not there could understand what happened.
4. Half the device's story is never spoken at all. When a vendor services the equipment, the record arrives as a PDF service report, often days later. No microphone captures that. A clinical scribe has no reason to handle it. An HTM one has to, because the surveyor will ask about the device, not about who held the screwdriver.
Who owns a work order an AI helped write?
The technician. Same as in medicine, where the clinician reviews and signs the scribe's draft. Service leaders have told us plainly that if the AI wrote it and the tech did not approve it, the tech did not mean it. They carry the liability, so they own the record. The AI removes the typing, not the judgment.
What should an AI scribe for HTM do?
If you are evaluating one, here is a short list to hold it to:
Works alone, hands busy. Short spoken notes during the job, not a recorded conversation.
Writes into your CMMS fields, in your shop's format. Not a paragraph you have to retype.
Sizes the note to the job. One line for the simple fix, full detail for the hard one.
Nothing goes in without the tech's yes. Review and approve, every time.
Handles vendor paperwork too. The service report PDF should become part of the same record.
Lets you type when talking is wrong. Occupied patient rooms exist. Voice cannot be the only way in.
Where does Leera AI fit?
This is what we build. With Leera AI, technicians describe the repair out loud while they do it, Leera AI writes the work order in the shop's own format, the technician reviews and approves it, and it syncs to your CMMS via API. For the vendor side, Leera AI turns vendor service report PDFs into real work orders: what was done, which parts, how long it took. That part is live today.
The VA just showed what happens when an organization decides documentation should not eat the end of everyone's day. Clinicians got that answer first. The bench is next.
FAQ
What is an AI scribe? An AI scribe listens to work as it happens and drafts the documentation, so the person doing the work reviews a note instead of writing one from scratch. In medicine it listens to the patient visit. In HTM it works from the technician's short spoken notes during a repair or PM.
Is there an AI scribe for biomedical technicians? Yes, though the category is young. Tools like Leera AI let technicians talk through a repair and turn that into a structured work order in the CMMS, which the technician reviews and approves before it is saved.
Does the VA's AI scribe contract cover biomedical engineering? No. According to the published coverage, it covers outpatient clinicians: primary care, behavioral health, rehab and medical and surgical specialists. HTM documentation is not part of it.
Can a clinical AI scribe be used for work orders? Not well. Clinical scribes are built to listen to a conversation and write a clinical note. A work order needs short spoken input from one person, CMMS fields like failure codes and parts, the shop's own format, and a place for vendor service reports.
Dima Okhrimchuk is the founder of Leera AI, which builds voice-first documentation for HTM teams. Leera AI is built in the field with HTM teams at major US health systems