The State of AI in Veterinary Medicine: Three Waves, and Where We Are Now

In the spring of 2024, Cornell's College of Veterinary Medicine hosted the first Symposium on Artificial Intelligence in Veterinary Medicine (SAVY), and the AVMA's coverage captured the mood of the moment: AI was "poised" to transform veterinary care. It was a fair summary of how the profession felt in 2024 — expectant, curious, a little wary. But it undersold how much had already happened, and it couldn't have anticipated how fast the next two years would move.

I've watched this from both sides — from the conference stage at SAVY and elsewhere, and from the deployment side, building AI that works inside veterinary clinics every day. What follows is a practitioner's read rather than a survey: my view of what has already happened, and where I think the next two years go. From that vantage point, AI's entry into veterinary medicine isn't one story. It's three overlapping waves, each with its own technology, its own adoption curve, and its own lesson. Knowing which wave a tool belongs to tells you more about how to evaluate it than any demo will.

Wave one: diagnostics and prediction (roughly 2019–2024)

The first wave predates the ChatGPT era by years, which surprises people outside the field. In October 2019, Mars Petcare and Antech launched RenalTech, a predictive tool built on two decades of records from more than 150,000 cats and 750,000 patient visits at Banfield hospitals, claiming better than 95% accuracy at predicting feline chronic kidney disease up to two years before conventional diagnosis. AI radiology services emerged in the same era, reading radiographs at a scale no teleradiology bench could match, and companies like ImpriMed applied machine learning to personalizing cancer treatment. By the first SAVY in April 2024, this wave had a full program's worth of mature work to show — kidney prediction, imaging triage, livestock analytics, even biometric identification by nose print.

Wave one established two things. First, that veterinary medicine had the raw material: decades of structured records, lab values, and images concentrated in large hospital groups and reference labs made genuinely predictive models possible. Second, the constraint that would define every subsequent wave — the model was rarely the hard part. The hard part was the data: fragmented across practice management systems, inconsistently recorded, and locked inside workflows built for a pre-AI clinic. Wave-one tools succeeded where someone had already paid the cost of assembling clean data at scale, which is why the breakthroughs came from Mars, the reference labs, and the universities rather than from the average three-doctor practice.

Wave two: documentation (roughly 2023–2025)

The second wave was the one the average practice actually felt, because it attacked the profession's most personal problem: time. When large language models became commercially usable in 2023, the first thing veterinary teams did with them was write — SOAP notes, discharge instructions, client emails. Dictation tools had been in clinics for years; what changed in 2023 was the leap from transcription to comprehension, from typing what you said to drafting the note for you. AI scribes and ambient documentation went from novelty to standard equipment faster than any veterinary technology I can think of.

The adoption data backs that up. A Digitail and AAHA survey of 3,968 veterinary professionals, fielded in December 2023 and January 2024 and later published in AJVR, found that 39.2% were already using AI tools at work — and of those, 69.5% used them daily or weekly. Read the number for exactly what it is: a convenience sample recruited through AAHA and Digitail channels, with a tech-adjacent skew, measuring individual professionals rather than practices. It is still the best peer-reviewed figure the profession has, which is why it sits in our verified statistics register with those limits attached. The top applications were imaging and radiology on one side and record-keeping, administrative work, and voice-to-text on the other: waves one and two, captured mid-handoff, barely a year after the LLM moment began. The same survey flagged what stood between the profession and broader adoption — 70.3% named reliability and accuracy as their top concern, and 53.9% named data security and privacy.

By SAVY 2.0 in May 2025 — 130 attendees in Ithaca, more than 80 joining virtually from 23 countries — the framing on stage had inverted from data scarcity to managing the flood of usable outputs, and the AVMA's follow-up coverage that September centered on documentation tools relieving administrative burden in real hospitals, including a large emergency group's scribe rollout. Scribes won adoption for a simple reason: they gave clinicians back an hour a day without asking anyone to change how they practice medicine. That became the template every later tool would be measured against.

Wave three: the front office (2026–2027)

The third wave is the one unfolding right now, and it moves AI out of the exam room entirely — onto the phone lines, the text threads, and the fax machine. This is the wave I work in, so discount my enthusiasm accordingly, but the shift is observable across the category: voice and text agents that answer clinic phones during the day and after hours, schedule appointments directly in the practice management system, run structured triage that escalates emergencies to a human on defined rules, chase down medical records from other clinics, and carry the follow-up messages that used to die on a front desk to-do list.

Three things converged to make this the 2026–2027 wave rather than a 2023 one. First, voice. A phone call is the hardest medium in the building — real-time, interruptible, emotionally loaded, often an emergency — and language models only recently crossed the threshold where a conversation with one is genuinely useful rather than a phone tree with better diction. Second, integration. The defining feature of a front-office agent is not that it talks; it's that it reads and writes — into the PIMS, the calendar, the client record, the fax line. That write-access layer, agentic rather than conversational, simply didn't exist in usable form during wave two, and an agent without it is an answering machine with a vocabulary. Third, trust. Clinics needed the scribe years to learn how to supervise AI — to review its output, correct it, and build instincts for where it fails — before they would let it near a live client interaction. Wave two was the apprenticeship for wave three.

The economics behind this wave are structural, not cyclical. Veterinary teams remain hard to hire and expensive to burn out; call volume doesn't respect staffing levels; and pet owners increasingly expect the on-demand responsiveness they get from every other service in their lives. The front desk is where all three pressures collide, which is why it's where AI is deploying fastest. Through 2027, I expect the frontier to move from answering to orchestrating: multi-channel continuity where a call becomes a text becomes a booking, per-department routing and after-hours rules, outbound workflows like record retrieval and follow-up sequences, and much deeper write-through into the PIMS. The industry is re-measuring itself as this happens — Digitail and AAHA opened a second industry-wide AI survey for 2026, and I'd wager the front office features in its results in a way it couldn't have in 2024.

Will AI replace veterinarians?

No — and in the rooms where this actually gets argued, almost nobody credible says it will. Notice what the three waves have in common: none of them practices medicine. Wave one surfaces predictions for a clinician to act on; wave two documents the medicine a clinician already performed; wave three handles the logistics around the visit. The pattern is consistent — AI absorbs the work around the medicine so the medicine stays human. Diagnosis, treatment decisions, and the veterinarian-client-patient relationship remain squarely with the DVM. The tools that have succeeded are the ones engineered around that boundary, with explicit escalation rules, human review, and a clear line between what the AI may do alone and what it must hand off. The profession's stance reflects the same balance: in the 2024 survey data, optimists outnumbered skeptics 43.1% to 36.9%, but the skeptics' concerns — reliability first, privacy second — are precisely the ones good deployments are engineered to answer.

What to ask before adopting anything

The category's maturity means the buying question has changed. In 2024 it was "is this real?" In 2026 it's "does this fit my clinic?" — and the answer depends on which wave you're buying from. For a wave-one diagnostic tool, ask what data it was trained on and how it was validated. For a wave-two scribe, ask how your team reviews and corrects its output. For a wave-three front-office agent, the evaluation that matters: Can it read and write to your actual PIMS, or does it create a second inbox your team has to reconcile? What exactly happens when a caller describes an emergency — who is alerted, in what order, on which channels? Where does client data live, who trains on it, and can you see and correct what the AI did, call by call? Vendors who answer those questions plainly are building for the clinic rather than the demo.

If you want the longer argument for what a front-office agent has to understand before it should be allowed to speak to a client, we wrote it up separately in From Systems of Record to Context-Aware Agents, and the call-level evidence behind the wave-three claims is in After-Hours at the General Practice.

Quick answers

How is AI used in veterinary medicine in 2026?

Four established categories: clinical decision support, predictive diagnostics, and imaging (established since roughly 2019); ambient documentation and AI scribes (mainstream since 2023–2025); front-office automation — phone answering, scheduling, triage, record retrieval, client follow-up (the 2026–2027 wave); and operations analytics on top of practice data.

When did veterinary practices actually start adopting AI?

Earlier than most assume. Predictive diagnostics like RenalTech launched in 2019, and by early 2024 a Digitail/AAHA survey of nearly 4,000 veterinary professionals found 39.2% already using AI tools at work — most of them daily or weekly.

Is there an AI that answers a veterinary clinic's phones?

Yes — purpose-built veterinary front-office agents (PupPilot among them) answer day and after-hours calls, book appointments in the PIMS, triage urgent cases to a human, retrieve records, and text with clients. The category emerged after the scribe wave and is maturing rapidly through 2026–2027.

What is SAVY?

The Symposium on Artificial Intelligence in Veterinary Medicine, hosted by Cornell's College of Veterinary Medicine — first held in April 2024, with SAVY 2.0 in May 2025 and subsequent editions announced at cornellaivet.org. It has become the field's best barometer for where veterinary AI actually stands.

About the author

Gary Peters is the CEO of PupPilot, the AI front office for veterinary clinics. He speaks on AI in veterinary medicine at industry events including SAVY, VMX, and the Veterinary Innovation Summit, is the author of a five-article AI series in AAHA Trends (March 2026), and is presenting four sessions on AI and client communication at AAHA Con 2026 in Portland. More of our work on where AI does and doesn't help on a clinic phone line is in the verified statistics register and the phone-first series.