Long-Term Futures: 2031–2036

All projections on this page carry explicit confidence labels based on the evidence base:

1. Foundation Models and Veterinary Clinical Reasoning

Where research stands now

General-purpose LLMs already pass veterinary licensing exams at high rates. A 2025 Frontiers in Veterinary Science study evaluated nine frontier models on 250 MCQs from a vet undergrad qualifying exam: ChatGPT 4.5 scored 90.4%, o1Pro scored 90.8% — well above the passing threshold, though performance degraded on harder reasoning tasks and image-based formats.^[1]

Stanford's VetLLM project (Pacific Symposium on Biocomputing 2024) demonstrated striking data efficiency: fine-tuning a 7B-parameter model on just 5,000 veterinary clinical notes produced diagnostic predictions surpassing supervised models trained on 100,000+ notes (F1 = 0.747 on Colorado State CVM clinical data).^[2] Early-stage and narrow, but the data-efficiency finding matters for a field that cannot train frontier-scale models from scratch.

Will there be a "Vet-GPT"?

RSP Almost certainly not as a standalone model from a dedicated vet AI startup — the economics do not support training frontier-scale models from scratch for a comparatively small market. More plausible: domain-adapted fine-tuning of existing foundation models on vet EHR data, diagnostic imaging datasets, and pharmacological databases. The field needs GPT-5 or its successor with veterinary RAG layers and species-specific fine-tunes, not its own foundation model.

Mars Petcare's network of approximately 2,000 US clinics plus the Biobank (accumulating genome sequences, metabolomes, and longitudinal EHR data from thousands of dogs and cats) represents the most plausible proprietary training data asset for a vet-specialized clinical reasoning model. Whether Mars licenses to an AI partner or builds in-house is an open strategic question.^[3]

By 2031–2036

RE Multimodal clinical AI co-pilots embedded in PIMS, surfacing differential diagnoses from SOAP notes, flagging abnormal labs, and suggesting treatment protocols with species-specific drug dosing. The bottleneck is EHR data interoperability and the profession's willingness to standardize diagnosis coding — currently a major barrier that Stanford's VetLLM was specifically designed to address.

2. Autonomous Diagnostic Agents

What's already happening — and the controversy

Three vendors — SignalPET's SignalSTAT, Vetology's AI Radiologist Report, and Antech's RapidRead — already deliver AI-generated radiograph interpretations to general practitioners without mandatory board-certified radiologist review, in a regulatory vacuum. SignalPET alone processes approximately 50,000 radiographs per week across 2,300+ clinics.^[4]

The ACVR/ECVDI 2025 position statement declared "currently, no commercially available AI products for veterinary diagnostic imaging meet the required standards for transparency, validation, or safety" — yet vendors continue operating. External validation published in JAVMA (2025) documented meaningful deficiencies in commercial vet radiology AI when tested on general-practice-sourced canine abdominal radiographs — the actual real-world deployment condition.^[5]

In human radiology: FDA has cleared 1,000+ AI imaging tools — every single one as clinician-assistive, not autonomous primary reader.^[6] Veterinary devices face no equivalent premarket review requirement.

By 2031–2036

RE Most likely outcome: tiered autonomy model. AI performs the primary screen; a radiologist (possibly remote) confirms abnormal findings. Full autonomous primary reads without human review are likely blocked by liability law and professional standards even if technical accuracy reaches required thresholds. Interestingly, the vet sector may actually leapfrog human medicine on establishing formal standards — the absence of existing regulation allows faster standardization from scratch if AAVSB or AVMA establish a SaMD-equivalent framework. Three conditions would need to be met for genuine full autonomy: a regulatory pathway requiring prospective validation studies, demonstrated AI accuracy surpassing board-certified radiologist performance on defined species/modality combinations under controlled conditions, and profession acceptance of liability frameworks for AI errors. None are close to resolved.

3. Robotic and AI-Assisted Surgery

Current state

Cornell CVM performed the first robotic-assisted gastropexy in the US using a da Vinci Si-e system supplied by R2 Surgical in 2024. Texas A&M Small Animal Teaching Hospital has faculty trained in robotic-assisted surgery. R2 Surgical is the primary company bridging repurposed da Vinci systems into veterinary medicine.

Research context: Johns Hopkins and Stanford integrated a vision-language model with da Vinci systems trained on 20 hours of surgical video to enable autonomous subtask performance (lifting tissue, needle positioning, wound suturing) in pig cadaver models — framed by NVIDIA as foreshadowing autonomous robotic surgery.^[7]

Vet-specific barrier: da Vinci systems cost $1.5–2M+ and are engineered for human anatomy. A 2024 PubMed review identified anatomical scaling and instrument miniaturization as the core unresolved technical barrier for routine small animal robotic surgery across the 2–60 kg patient size range.^[8]

By 2031–2036

RE Robotic surgical assistance commercially available in large university referral hospitals and well-capitalized specialty practices. Unlikely to be routine in general practice. Economic case improves as second-generation systems designed with veterinary anatomy in mind enter the market. AI-guided autonomy for specific subtasks (suturing, knot-tying) during laparoscopic procedures is plausible within this window, but full autonomous surgery is not.

S If foundation model application to surgical video continues at its current trajectory, AI surgical coaches — systems providing real-time guidance and flagging error-prone maneuvers during human-performed surgery — could be the first widely adopted "autonomous" surgical AI in veterinary medicine. This would avoid the liability problem of replacing the surgeon while delivering meaningful safety improvements.

4. Precision Veterinary Medicine: Genomics + Clinical AI

The most robustly funded and research-backed long-term trajectory in this space.

The Mars Petcare Biobank (10-year commitment, up to 20,000 pets, integrating genome sequences, fecal metagenomes, EMRs, clinicopathology, and activity data) has, three years in with 4,500+ enrolled pets, already linked SLAMF1 variants to canine atopic dermatitis. Full genome datasets are being deposited in NCBI Sequence Read Archive as open access.

The Dog Aging Project (NIH-funded): 976 companion dogs in a precision multi-omics cohort (metabolomics, microbiome, epigenomics, immunophenotyping, clinical pathology) between 2021 and 2024, with ongoing annual follow-up. 2025 GeroScience publication describes the cohort design and open data release via Terra platform.^[9] A 2025 Aging Cell paper identified protein catabolite biomarkers of aging physiology from the metabolomic data.^[10]

Embark's 2025 integration with Innovative Pet Lab brings microbiome, immune biomarkers, and genomic data together for longitudinal wellness tracking.

By 2031–2036

RSP Genetic risk stratification routine at the time of adoption — flagging breed and individual-level predispositions to cardiac disease, cancers, metabolic disorders, and drug metabolism variants. The data infrastructure is being built now. The hard problem is EHR integration: whether genomic risk profiles flow into clinical decision support and trigger earlier screening or preventive interventions.

RE Multi-omics panels (genomics + microbiome snapshot + metabolite biomarkers) available as reference lab add-ons for wellness visits. Dog Aging Project longitudinal findings will directly inform which metabolites and epigenetic markers are clinically actionable, likely by early 2030s.

5. Telemedicine and Decentralized Diagnostic AI

The vet telehealth market was $306.7 million in 2024, projected to approximately $921 million by 2030. Over 60% of vet practices offered some virtual care in 2026. Consumer-facing AI tools include TTcare (1.4M+ scans, claimed 95% accuracy for eye, skin, dental, and gait analysis), DogMD, Vet-AI's Joii 165-algorithm symptom checker, and PetHub 24/7 telehealth.

By 2031–2036

RE The "first contact" model shifts substantially: owners capture standardized video, wearable data, and potentially home urinalysis or blood glucose readings that feed AI triage routing to either asynchronous specialist review, telehealth consultation, or an in-clinic visit. The in-clinic visit becomes reserved for hands-on examination, advanced imaging, and procedures. This trajectory already defines human primary care; veterinary medicine follows with a lag.

S Owners in 2033–2036 carry handheld microfluidic devices running basic blood panels with AI interpreting results in real time. Prototype technology exists for human medicine (Theranos is the cautionary tale on overreach; Scanwell Health and Sight Diagnostics are more disciplined current players). Veterinary adaptation would follow, but validation and regulatory clarity must precede consumer deployment.

6. Owner-Facing AI and the Vet Business Model

AI pet symptom checkers represent a $342 million market in 2024 projected to reach $1.18 billion by 2033 at 14.8% CAGR.^[11] This is already influencing when owners seek care.

By 2031–2036

RE Bifurcated impact on visit volume. Low-acuity visits (mild GI upset, minor lacerations, routine medication refills) increasingly handled via AI-assisted telehealth, reducing foot traffic. However, AI may simultaneously increase net veterinary visits by lowering the friction barrier for early intervention on conditions owners would previously have "watched at home" until severe.

Structural risk to the practice model is NOT about AI replacing veterinarians — it is about AI enabling unbundling of services. If AI handles triage remotely, drug dispensing shifts online, and basic lab work shifts to at-home kits, the brick-and-mortar practice is left with procedures, imaging, and complex case management — highest value but lowest volume.

RE Corporate groups (Mars Veterinary Health, JAB/IVC Evidensia, NVA) that own large clinic networks absorb this better because they can vertically integrate telemedicine, diagnostics, and AI tools across their network. Independents face margin compression. Corporate practices grew from approximately 10% of the US market in 2017 to approximately 30% by 2025; AI adoption advantage likely accelerates this trend.^[12]

7. Regulatory Evolution

Current state: a gap, not a framework. AAVSB AI guidance whitepaper (2025), AVMA Task Force (convened late 2025), ACVR position statement (June 2025) — professional society positions, not enforceable regulations.

FDA CVM has expressed interest in extending Software as a Medical Device (SaMD) principles to veterinary applications but no formal pathway exists yet.^[13]

Most plausible 5–10 year regulatory trajectory

RE Three likely developments, roughly in order:

1. AVMA/AAVSB establish a voluntary certification standard for veterinary AI diagnostic software — a vet-specific SaMD framework equivalent adopted by state boards as a condition of practice compliance. No Congressional action required.
2. FDA extends CVM oversight to software making diagnostic claims about animals, likely triggered by a patient harm event generating political pressure. EU AI Act application to vet AI tools sold in Europe may push international harmonization faster than US domestic pressure alone.
3. State practice acts get tested: the legal fiction that AI-generated radiology "findings" are not "diagnoses" is eventually litigated. A malpractice case where a misread AI report causes patient harm and the vendor successfully disclaims responsibility would force legislatures to clarify.

S International standardization: if major markets (EU, UK, Australia, Canada) align on minimum validation requirements for veterinary diagnostic software, analogous to medical device harmonization, this could drive US standards faster than domestic legislative processes.

8. The Non-Obvious Long-Term Implication: Inversion of Clinical Value

The most important long-term implication for veterinary hospitals as businesses is not about any single technology. It is about the inversion of where clinical value is generated.

Today, veterinary hospitals generate most diagnostic value at the moment of examination and testing. The veterinarian sees the animal, interprets data, and renders judgment — all within the same encounter, the same location.

The 5–10 year AI trajectory effectively disaggregates this:

• Genomic risk stratification generates clinical intelligence before the animal is sick
• Continuous wearable monitoring generates data between visits
• AI triage handles first contact
• Remote specialists review imaging asynchronously
• The in-clinic encounter becomes one node in a longitudinal care relationship, not the primary site of intelligence generation

This is a fundamentally different business model — closer to primary care medicine's aspiration for "longitudinal health management" than the episodic, transactional structure currently defining most veterinary practice.

RE Practices positioned well in 2031–2036 are not those with the best AI diagnostic tool. They are those that have built the client relationship and data infrastructure to serve as the trusted hub of longitudinal records. Corporatized networks are explicitly building toward this — Mars's Biobank is not just a research asset, it is a data moat.

The structural threat is not AI replacing veterinarians. It is a small number of vertically integrated networks controlling AI, data, diagnostics, pharmacy, and client relationships simultaneously — while independent practices survive by differentiating on what AI cannot yet replace: species specialization, complex case continuity, client trust, and the irreducible value of a skilled clinician's hands on a sick animal.