The Tool Doesn’t Do the Task. It Takes the Reps.
Robert Grand · Battalion Chief who still runs calls
Every pitch for AI in this service sells the same promise. It handles the boring part so your people can spend their attention where it counts. The report narrative, the triage note, the records grind. Hand it off, free up the crew, everybody wins. It is a clean story and most of it is even true.
We’re handing off the reps that built the judgment.
That is the part the slide never shows. The work being restructured isn’t the typing, it’s the reasoning, and the reasoning is how the job teaches itself.
What the Medical Research Just Named
Medicine got there first, and this year the literature put a word on it. A scoping review of AI in clinical practice found that overreliance erodes physicians’ independent reasoning, that performance drops measurably when the AI is taken away, and that the people most exposed are the trainees, who risk what the research now calls “never-skilling,” failing to build the foundational competence because they leaned on the tool before they ever did the work cold (Artificial intelligence in medicine: a scoping review of the risk of deskilling, https://www.esmorwd.org/article/S2949-8201(26)00012-3/fulltext).
Imaging work goes further. When people repeatedly offload the thinking to a machine, the prefrontal cortex gets less active during the task, and independent reasoning capacity measurably declines (Deskilling dilemma: brain over automation, https://www.frontiersin.org/journals/medicine/articles/10.3389/fmed.2026.1765692/full). The same concern is showing up in prehospital telemedicine, where the ethics of leaning on decision support in the back of a medic unit are now an open question, not a settled one (AI support in prehospital telemedicine: ethical considerations, https://pmc.ncbi.nlm.nih.gov/articles/PMC12833132/).
Read those three together and the pattern is plain. The tool doesn’t just do the task. It quietly takes the reps that the task used to build, and it does it without anyone noticing, because the output still looks fine.
The Medic Two Years In Who Never Started Cold
Picture her. Two years on the job, sharp, well liked, good hands on a patient. She has written a hundred ePCRs and the AI has structured every one of them. She can read a draft, fix a line, catch a wrong med. What she has never done is sit in front of a blank screen and build a messy call into a defensible patient story from nothing.
That blank-screen skill is the one that holds up in a deposition two years later. It is the one she needs on the call where the tablet is dead, the connection is gone, and the patient in front of her is circling. The tool that made her fast also made sure the rep that builds that skill never happened.
Same pattern on the officer side. A newer company officer leaning on AI decision support builds speed, not size-up. He gets good at confirming what the model already told him and never gets the reps that build the gut you only earn by being wrong a few times and sitting with it. The research line that should stop you cold is the one about performance dropping when the AI is removed. In our world the AI gets removed at the worst possible moment, on the bad call, in the dead zone, when the system is down. What’s left is whatever the person actually built.
I’ll be honest about my own habit here. I write about this and I still catch myself reaching for the draft button before I’ve thought the call all the way through on my own. The pull is real. The efficiency is real. That’s exactly why this is hard to see and harder to fix.
The Rep Was Never the Typing
Here is the reframe the vendor will never offer you. When the model drafts the narrative, the thing it replaces isn’t the keystrokes. It’s the mental act of building a story out of a scene, deciding what matters and what doesn’t, holding the timeline together and committing to it. That act is how judgment gets built, one call at a time.
That work is not overhead. It’s the apprenticeship hiding inside the paperwork.
For a hundred years this job taught itself, because the reps were unavoidable. There was no other way to get the work done, so you got good by doing it. AI breaks that quietly. It removes the rep while leaving the work done, and competence that used to form for free now has to be built on purpose, or it doesn’t form at all. That changes the whole shape of how we bring people up, and most of us haven’t noticed the floor moving.
What “Load-Sharing” Actually Means When the System Goes Down
The vendors call it load-sharing, and they’re not lying. The tool genuinely shares the load. The question they don’t ask is which load, because there are two kinds and they look identical on a demo.
One kind is the load you’re glad to give away: the formatting, the field population, the boilerplate that never built anything in anyone. The other kind is the load that was secretly the training: the reasoning, the narrative, the call you have to make about what goes in and what stays out. Hand off the first and you’ve bought time. Hand off the second and you’ve sold the apprenticeship to buy the time, and you won’t see the bill until a newer provider hits the call that needs the skill nobody let her build.
So load-sharing in our service has to mean something more specific than it does in an office. It means deciding, on purpose, which reps you protect. It means drills that make crews work cold, sign-offs that test reasoning and not just output, and an honest answer to who owns skill maintenance now that the daily job no longer maintains it. None of that is anti-tool. A service that does this gets a workforce that’s fast with AI and sound without it, which is the only combination that survives a hard call on a bad day.
The Skill That Doesn’t Show Up on Any Dashboard
The trap is that a rep quietly not forming is invisible. There’s no alarm, no metric, no budget line for the judgment a medic didn’t build because the tool built the chart for her. Short-staffing makes it worse, because when you’re running lean and the tool gets the narrative done in a third of the time, telling crews to do it the hard way sounds like a luxury nobody can afford. You only find out what wasn’t there on the day you needed it.
The efficiency is real and it’s not going away. The move isn’t to refuse the tool. The move is to decide, on purpose, which forms of competence you refuse to let it absorb, and then to build the training that protects them. A service that does that keeps its judgment and its speed both. A service that lets efficiency make the call by default will field a cohort five years in that’s fluent with the system and thin underneath it, and it won’t see the gap until the day the screen goes dark and the skill isn’t there.
Robert Grand is a Battalion Chief at Eugene Springfield Fire with 24 years of service. He writes Frontline Intelligence, a newsletter on operational doctrine, technology, and leadership in Fire & EMS.
If this landed, share it with someone in Fire & EMS who needs to hear it.
Subscribe
From the floor, not the vendor booth. Two times a week.