From Systems of Record to Context-Aware Agents
Rethinking front-office AI in veterinary medicine.
Veterinary AI is often discussed through the lens of diagnostics and scribing. Those are important categories, but they are not the only places where AI will change how veterinary medicine works.
One of the largest near-term opportunities is the front office.
This piece is based on a talk Gary Peters, PupPilot's CEO and co-founder, gave at the Cornell Symposium on AI in Veterinary Medicine. He studied Applied Mathematics and Scientific Computing at Vanderbilt — a program now relabeled as AI — and did his graduate work in AI at Carnegie Mellon. Since then, he has spent years in the Stanford startup ecosystem, through programs like StartX and Cardinal Ventures.
His thesis at Cornell was simple: the front office is where administrative AI can unlock medical impact. That does not mean AI replaces veterinarians. It does not mean AI diagnoses patients, signs medical records, or assumes clinical liability. The opportunity is different. It sits in the operational layer around care: the calls, texts, scheduling loops, reminders, follow-ups, triage routing, and client communication that determine whether the care plan actually happens.
This article is based on Gary Peters' presentation at the Cornell Symposium on AI in Veterinary Medicine. The session recording will be added here once Cornell publishes it.
When does theory become practice?
The thread running through Gary's career is the bridge between the theoretical and the applied — the move from studying AI in the abstract to putting it to work in the real world. So the talk opened with a question: when does a new technology actually cross that line into everyday practice?
His answer was that two things have to change. They are the reason and rhyme behind everything that follows, and the rest of this piece works through both — and what they unlock for the veterinary front office.
- 1
From "what" to "why"
Software stops simply moving information around and starts understanding why it matters.
- 2
Who adapts to whom
People stop bending to the software, and the software starts adapting to them.
The shift from "what" to "why"
The last generation of front-office software largely solved the "what."
Can the clinic make and receive phone calls? Can clients send text messages? Can voicemails be stored? Can reminders be sent? Can appointment requests be collected? Can messages appear in one inbox?
Those tools were useful. They helped clinics move communication out of sticky notes, paper call logs, and disconnected systems. But they still depended on a person to supply the context.
A client sends a text. A human reads it. A voicemail comes in. A human interprets it. A pet owner asks for an appointment. A human decides what kind. A client asks a medication question. A human looks in the chart. A post-op client reports a change. A human decides whether to escalate.
The software handled the channel. The human still had to understand the patient, the plan, the clinic's rules, the urgency, and the next step.
That is the shift AI makes possible. The next generation of veterinary front-office technology will not be defined by generic task automation. It will be defined by context. Why this patient? Why this instruction? Why this urgency? Why this next step? Why this escalation?
That is the difference between a communication tool and a context-aware agent.
AI adapts to the user, or the user adapts to AI
The second shift is subtler. It is about who adapts to whom.
Scribing is a strong example of AI adapting to the user.
The veterinarian still sees one patient at a time. The appointment flow looks mostly the same. The doctor talks with the client, performs the exam, makes the medical decision, and the AI reduces the documentation burden afterward. That is valuable. It saves time inside the existing workflow.
Front-office AI is different. It changes the shape of the work itself.
A human team is constrained by sequential communication. One person can answer one call, send one text, or make one follow-up call at a time. AI can operate concurrently across inbound and outbound communication.
That means the clinic does not simply do the same work slightly faster. It gains a new kind of capacity. The front office can answer, route, follow up, schedule, remind, summarize, and escalate at a scale that human teams could not previously deliver.
Why older tools stayed shallow
Historically, software was expensive to build. To justify that cost, many companies built broad tools that could serve many markets at once: veterinary clinics, dental offices, chiropractic offices, medical practices, home services, and other appointment-based businesses.
That pushed products toward breadth. A phone system could serve every market. A texting inbox could serve every market. A voicemail tool could serve every market. A generic scheduler could serve every market.
But breadth usually comes at the cost of depth. A general-purpose communication tool can receive a message, but it does not deeply understand veterinary medicine. It does not know whether a patient is post-op, whether a medication question relates to a documented treatment plan, whether a recheck is overdue, whether a symptom should trigger escalation, or whether a scheduling request requires a specific appointment type.
AI changes the economics and the product opportunity. Large language models can understand messy client communication in ways older software could not. AI coding agents and modern development tools also make it faster and less expensive to build deep vertical software than it used to be. That makes it possible to build specifically for veterinary workflows instead of stretching one shallow workflow across many industries.
The result is a shift from tools that move messages around to agents that understand what the message means and what should happen next.
The front-office gap
Most veterinary clinics already have the data needed to answer many client questions. It lives in the practice information management system. The PIMS contains clients, patients, history, medications, lab results, treatment plans, appointments, invoices, and medical notes.
But that data often does not travel into the live conversation.
The phone system does not know the full patient context. The voicemail box does not understand the treatment plan. The texting inbox does not know whether the issue is urgent. The scheduler does not always understand clinical reason-for-visit rules. The reminder system does not explain why the care matters.
That is the front-office gap.
The system of record stores what happened. The front office handles what is happening now. The problem is that those layers have historically been disconnected. A context-aware AI layer connects the data to the conversation.
The anatomy of an AI front office
A true AI front office is more than a chatbot or answering service. It needs to move from conversation to action. There are five layers between a ringing phone and a documented outcome.
- 1
Voice interface
The AI needs to handle natural conversation, interruption, speech input, and speech output.
- 2
Caller and patient identification
The system needs to match the caller to the correct client and patient record.
- 3
Context retrieval
The AI needs to retrieve relevant information from the PIMS, chart, documented treatment plan, and clinic knowledge base.
- 4
Action layer
The system needs to schedule, summarize, route, follow up, or prepare the next step.
- 5
Escalation and writeback
When a human needs to take over, the handoff should include a structured summary. When the workflow is complete, the record should be updated where appropriate.
The principle is simple: retrieval over invention. The AI should not invent the medical plan. It should retrieve, explain, and operationalize the plan the care team already documented.
Inbound communication is the first unlock
Inbound communication is where the pain is easiest to feel.
The phone rings while the receptionist is checking in a client. Another call comes in during discharge. A voicemail arrives during surgery drop-off. A client texts about a refill. Someone wants a wellness appointment. Someone else is worried their dog ate something. Everything enters the same bottleneck.
That is why inbound AI support creates immediate relief. It can answer routine questions, schedule what can be scheduled, collect the right information, and escalate what should not wait.
That is why the front office is such a strong AI use case. The majority of the communication burden is not complex medicine. It is routine operational work that follows known clinic rules. The AI can handle the routine work so the team is not constantly interrupted by tasks that do not require human judgment.
Different calls need different workflows
Not every call should be handled the same way.
A general FAQ can often be answered directly from clinic-approved information. A scheduling request may require live calendar access and appointment rules. A medication or refill question may require chart context and a documented treatment plan. A post-op question may require surgical notes, discharge instructions, and escalation logic. An emergency call should be escalated conservatively.
This is where context matters. A generic system can take a message. A context-aware agent can understand what type of workflow the conversation belongs to and what the safest next step should be.
Outbound communication is the bigger unlock
Inbound communication reduces chaos. Outbound communication changes what clinics can reliably do.
Most clinics already know what good follow-up should look like. A post-op patient should be checked on. A chronic-care patient should return for monitoring. A client should understand the medication instructions. A pet overdue for preventive care should be contacted. A patient with a new diagnosis should not disappear into a reminder queue.
The problem is not that clinics do not care. The problem is that they are already overloaded. When the front desk is barely keeping up with calls, texts, check-ins, checkouts, refill requests, and appointment scheduling, proactive outreach becomes inconsistent. It happens when someone has time, and often no one has time.
AI changes that because outbound work can happen in parallel. One staff member can make one call at a time. A context-aware AI system can run thousands of outreach workflows at once. That does not replace medical judgment. It scales the coordination around medical judgment.
For example, an AI agent can call two days after surgery, ask about appetite, incision appearance, and medication use, then escalate a concerning response to the doctor with a structured summary. The veterinarian still makes the medical decision. The AI makes sure the follow-up actually happens. That is the care-coordination opportunity.
What this means for care
The most exciting part of AI front-office support is not that it answers calls faster. It is that it helps clinics deliver the care they already wanted to deliver.
A veterinarian may recommend a recheck, monitoring bloodwork, medication follow-up, post-op assessment, chronic-care reminder, or preventive-care visit. But the clinic still has to make that happen through staff bandwidth, reminder systems, client memory, and manual follow-through. That is where care plans often break down.
Context-aware AI can help close the gap between the recommendation and the completed next step. It can remind the client. It can explain why the recheck matters. It can answer basic questions from the documented plan. It can schedule the appointment. It can check in after the visit. It can escalate if something sounds wrong. It can summarize the interaction for the care team.
The care team still owns the medicine. The AI coordinates the logistics around it.
Safety: context-aware does not mean fully autonomous
Context-aware AI should not be confused with fully autonomous medical decision-making. A safe AI front office needs clear boundaries:
- Clinic-controlled protocols
- Retrieval over invention
- Conservative triage
- Human escalation on ambiguity
- Structured conversation summaries
- Writeback where appropriate
- Clear separation between documented care instructions and new medical judgment
For medical ambiguity, the system should bias toward escalation. It is better to escalate a concern unnecessarily than to miss something that needed attention. A false-positive escalation is safer than a false-negative reassurance.
That is why live transfer and after-hours routing matter. During business hours, urgent or ambiguous issues should route to the clinic team. After hours, they should follow the clinic's approved pathway, such as an emergency hospital, pet poison control, an on-call clinician, or another designated resource.
The AI should be helpful, bounded, and conservative.
From recordkeeping to coordination
Veterinary software has historically been built around systems of record. Those systems are essential. They store the facts of care. But the next generation of veterinary AI will be built around coordination.
Yesterday's systems recorded what happened. Next-generation agents coordinate what happens next.
That is the real shift from "what" to "why." Not communication tools that simply move messages around, but context-aware agents that understand the patient, the plan, the protocol, the moment, and the safest next step.