2 Jun 2026
AI has the power to transform clinical workflows and administrative tasks through sophisticated agents and voice technology. While human oversight remains essential, successful integration of these tools directly into existing systems can help unlock new operational efficiencies…
Oli Viner
Job Title
Image: gorodenkoff / iStock
It seems impossible to move at the moment without encountering someone talking about AI.
Rather unusually, veterinary medicine – often considered something of a backwater for technological innovation – is actually surprisingly far along in the adoption curve.
This is in part because we have fewer regulatory hurdles compared to our medical and dental counterparts, but is also accelerated by the general awareness of AI tools for personal use, as well as the increasing availability of tools within existing solutions such as the PMS or diagnostic equipment.
Much like any advancement in technology, however, its benefit to end users is pretty closely coupled to their ability to use it meaningfully within their day to day work.
No matter how good the AI tool is, if it’s totally siloed from the PMS, 16 clicks and two lots of multi-factor authentication away it isn’t really going to dramatically change your working life. The best tools can gather context from what you are doing, and can be accessed easily and simply from your existing tool set. They also write data back to where you need it without cumbersome additional clicks and streamline workflows.
Veterinary professionals still remain liable for decisions they make, even if they are augmented by the use of AI. As such, most AI remains an assistant rather than a total autopilot. This does mean that there are some limitations to full workflow automation – some workflows that feel like they could be automated entirely may still require a pause and reflect for a human along the way.
Over time, it is possible that these workflows could be automated based on feedback and iteration from those manual steps. Equally, there may be exit points where a workflow is automated unless a certain threshold is met, at which point it exits for human intervention.
Defining those thresholds, and also defining what level of automation is appropriate, is still a grey area in the veterinary field, so maintaining some direct human involvement in every clinical workflow is standard as of the time of writing.
You may have already heard about AI agents, but for those who haven’t, agents are AI “programs” that can run and take action autonomously – that is, they do not require a human to perform an action.
By way of an example, if you asked ChatGPT to help suggest a new pair of trainers, a “non-agentic” response could give you a table, detailing pros/cons of all sorts of trainers for you to act on. An agent in this scenario could instead be tasked with actually buying them – namely, you empower the agent to find and source the best trainers based on your criteria. After setting it off, the next you are aware of it is a box of trainers at your doorstep. Agents can run asynchronously (you set them off and leave them to work) and can also work in teams or swarms.
Agents often use a lot of what is called “chain of thought” reasoning. While this can occur with non-agentic use in “deep thinking” modes, it is more common when agents are involved. What this means is that the agent will make a plan, analyse its own plan, adjust the plan, then start going through items one by one. At each step, it will evaluate the performance of that step against expectations and can adjust accordingly.
As you can imagine, agents can be immensely powerful, and are likely to be the future of computer use for most of us. They can be fantastic for optimising workflows and increasing human efficiency. However, they come with the downside that they perform actions without human oversight, and in a regulated world such as veterinary medicine this comes with some concerns and challenges.
However, this does not stop us using agents for things that are non-clinical. An example might be your marketing. A tool such as Claude can be given “skills” to feed it with context about your business, as well as how to run an effective marketing campaign.
You can then get Claude agents to generate new marketing content in your tone of voice, with your design, and then create advertising campaigns on Meta/Google. You can use agents to monitor the responses, and dynamically alter the content to ensure those adverts getting the most attention are biased towards.
Anthropic (the company behind Claude) is worth somewhere in the region of US$400 billion and up until fairly recently their entire marketing department was one person using Claude. It really is getting ever easier to use off the shelf AI tools like Claude to automate non-clinical workflows in their entirety.
Beyond standard AI agents, there is the possibility to use AI voice agents. These are natural language tools that understand speech. While some of us will have PTSD from using cinema booking phone lines in the 90s, modern AI voice agents are something else. They have near zero latency, can add pauses, tone, even filler words (“ums” and “ers”) as appropriate.
Voice agents provide the possibility of removing workflows that were previously the domain of humans. Non-clinical pathways such as new client registration and appointment bookings can be performed by a voice agent connected to (or native within) your PMS. They can be trained with your specific business processes, as well as your company tone of voice.
Evidence suggests that particularly when these are made “opt in” for clients, in circumstances such as out of hours or where the humans are too busy to answer, they are very well received and engaged with.
Voice agents may also be able to help us with clinical/para-clinical workflows. For example, confirming and then collecting information before a surgery can be done by outbound voice agents. They can also perform follow ups after surgery, gathering important information for review by a human. These are being used with considerable success in the human medical field.
Workflows are usually defined as a sequence of repeatable discrete tasks. Certain actions in a vet clinic may not be workflows according to that definition, but could be if we embrace new ways of working. Some typical workflows, however, are the consultation workflow, product/prescription requests and surgical flows.
A consultation can be thought of as connected discrete tasks:
AI scribes, such as Heidi, VetRec, CoVet and others, are in use in many veterinary clinics (and in some cases human hospitals). While most started as tools that simply listened to the consult and transcribed the output, many now offer true workflow capabilities, such as:
All the aforementioned can potentially be achieved by simply engaging the scribe at the start of the consult, and allowing it to work. While the vet will need to read and check the output, as well as confirm the actions and invoicing are accurate, the time required to do all of these tasks can be radically reduced. On top of this there can be additional benefits, such as having a full transcript in the case of a client complaint or for training.
We are all aware of how time consuming product requests can be. The discrete tasks here are (at the very least):
There are many branches on this workflow; requiring clarification from owners, engagement with third parties, issues with stock, as well as the possibility that the request may be denied and the pet requires a physical exam.
Under your care guidelines for dispensing of medications you can also add some complications that weren’t there before, such as requiring a prescription for an appropriate parasiticide treatment to already be on file (or require a re-exam).
AI can be of assistance here in most of the steps apart from the clinical authorisation one. The system can identify requests from conversations (digital or even voice) and then start a workflow to follow that item through the steps of acquiring payment, allocating the product and informing the owner. These steps can significantly reduce administrative overheads.
Animals coming into the hospital for a surgery or procedure also have a multistep workflow that can be heavily improved with tech. The discrete tasks here are:
AI systems can combine these steps into streamlined workflows, bringing consent capture forward to asynchronous communications and updating the PMS on the fly. Over time we may even see wearables/trackers that can follow animals through the physical building. Managing surgical follow-ups is also time consuming and something that AI can enhance, setting differentiated schedules (and potentially mediums) for that engagement.
On the non-clinical side, most managers and clinical teams struggle with rota and diary management. This is often complicated by poor PMS support for viewing rotas easily, and a disconnect between HR systems and the rota as it exists in practice – sometimes two very separate things. Another aspect is fairness – it can be difficult not to feel hard done by at certain shift changes.
AI can monitor whatever source of truth you define (for example, the HR system), and off the back of changes there, implement updates to the clinical rotas. This can include managing holidays, sickness, or CPD requests in a way that is fair and simple. Even to the granular level of managing availability in the PMS, AI can make changes to the diary even on the day in a way that removes any bias.
Oli Viner is co-founder of Hello Vet – one of the UK’s fastest growing veterinary startups, creating clinics where he’d want his kids to work and pets be treated. He’s possibly the world’s only veterinary surgeon and full stack engineer, with extensive experience as a veterinary surgeon as well as business owner.