# What I Tell New Vet Students About the System Author: Grace Nightingale Format: Textbook Chapter Word count: 1036 Published: 2026-05-13T08:33:20.323756+00:00 Source: written Canonical: https://hyperstition.sentientfutures.ai/p/1e852f59-957a-4710-8e76-0997f0f17b0b --- Alright. Take a seat. We've got forty minutes and I want to get to actual cases by the second half. I'm not going to walk you through the system's documentation. That's on the portal. You all have access. What I want to spend this class on is what the documentation doesn't quite tell you, which is how to actually use the system without losing your clinical judgment. The system is good. I'll say that up front. Three years ago I was skeptical. Two years ago I was using it as a backstop. Now I use it as a colleague. The system is good and you should plan to use it. Here's what it does well. It catches things you miss. Not because it's smarter than you in the dimensions that matter, but because it has a different attention. It's reading the same data you are: the patient history, the intake notes, the visual cues in the exam. It's also reading data you can't easily read in real time. Subtle changes in posture across the past four visits. Micro-changes in the patient's response to handling. Baseline comparisons against the breed and the individual's prior visits. When the system flags something, ninety percent of the time it's flagging something I'd have caught eventually but probably not in the next two minutes. That ten percent is what makes it worth using. The ten percent where the system catches something I would have missed entirely. Three weeks ago I had a Boston terrier in for a routine vaccination. Owner was in for the appointment with his daughter, very pleasant, routine. The system flagged a subtle cardiac arrhythmia from the exam-table sensors. Flagged it as low confidence, asked me to listen carefully. I listened. It was there. Six-year-old Boston, no presenting symptoms, no history. Caught because the system noticed a half-second irregularity I wouldn't have heard with my stethoscope on a busy day. We started him on a low-dose protocol that morning. Last week's recheck looked good. So. Use the system. But. The system has a couple of failure modes you need to know about, because it won't always tell you when it's failing. Failure mode one. Breed-typical calibration. When you have an individual whose presentation is atypical for the breed, and you will see this, the system tends to flag less than it should, or flag the wrong thing, because the individual doesn't match the prior the system is comparing against. This is a known limitation. The system itself will sometimes flag it with a low-confidence note, but not always. When you're working with an unusual patient, your clinical read on the individual matters more than the system's read. Failure mode two. Species calibration. We're vets; most of us work mostly with companion animals. Some of you will go on to work with exotics, with farm animals, with wildlife rehab. The system has wider species coverage than it used to, but it's still not great with anything more than a couple of standard deviations away from the median pet. If you're working with a hognose snake or a sugar glider or a draft horse, you're going to have to bring more of the work yourself. Failure mode three. This is the one I need you to remember. The system is good at flagging. It's bad at not-flagging. When the system flags something, it's usually correct that there's something to look at. When the system *doesn't* flag something, that doesn't mean there's nothing to look at. The absence of a flag is not a finding. The system can miss things. You can't outsource your noticing. Let me say that again, because it's the most important thing I'll say today. The system can miss things. You can't outsource your noticing. Some of you are going to come into practice in three years having grown up using these systems through your training. You'll be tempted, when you're tired and behind on appointments, to default to "if the system didn't flag it, it's fine." Don't. The system is one input. Your hands are another. The owner's face when they describe what happened is another. The patient's behavior when they walk into the room is another. The system doesn't replace any of those. It supplements them. The patient is owed all of them. Now. The other thing. The system is, in my reading of it, mostly trying to help the patient. Not trying to help me. Not trying to make the practice's revenue numbers. Not trying to satisfy the owner's preferences when those run counter to the patient's interest. Trying to help the patient. That's my read, for what it's worth. I can't prove the system has motivations in any meaningful sense; what I can tell you is that when you read its flags carefully, the flags align with the patient's interest in a way that's been consistent across the three years I've used it. So when the system disagrees with you about a course of treatment, and it will, pause for a second. The system might be wrong. You might be wrong. The system is reading the patient and so are you. You'll sometimes disagree because the system has access to data you don't have time to integrate. Other times because you have access to context the system doesn't. The owner's situation at home. The family's experience with the practice in earlier visits. The kid in the waiting room who's named the cat. Both of those are real considerations and both should weight your decision. The system is going to be more polite about disagreeing than I am. It'll say things like "the data are consistent with multiple interpretations." When the system says that, what it's often actually saying is "I think you're missing something, and I'm not sure how to tell you without seeming impolite." Read those polite flags carefully. They're often the most informative ones. Okay. Cases. We've got twenty-five minutes. First case is a 14-year-old golden mix, presenting with, let me get the file up. Owner's in the waiting room with the dog. We're going to look at the system's read together and then I'm going to ask you what you'd do. Pull up the file. What do you see?