15 October 2025

I thought voice agents were a problem looking for a solution. I’ve changed my mind

Comment Technology

It’s an exciting time for the next iteration of primary care AI wars.


The scribe wars have dominated the last 12 months and we are now entering the AI receptionist era.

Go to any conference and you will find a bevy of organisations expanding from scribe offerings to the “administrative efficiency” of AI reception staff.

We’ve been using Heidi Health voice agents over the last month and it’s been an enjoyable trial.

Initially (as a practice owner and GP) I just wanted to have a go to see what all the excitement was about. But as the technology beds itself in I’m surprised at how many use cases can be spun out with unlimited time and unlimited resources.

So, first off, what’s the technology?

Voice agents offer both outbound and inbound voice agents spun out from a dashboard. You can create multiple agents at will and control the knowledge base. Each agent is a walled garden.

Having that control over what the agent is allowed to know means you can actively iterate inside the product with live testing as you spin up each agent.

What does this mean for the end consumer? A friendly human sounding voice which can perform certain tasks depending on how you configure them.

Sitting around late on a Friday night testing agents and linking them up to our phone system live is an amazing representation of how simple IT has become. If I had to rate my technical ability it would be “can screenshot on a Mac but can’t use a video editor” and I found it easy.

Interestingly, the most challenging step is defining your outcomes and building a knowledge base in a way that reliably reproduces what you want. Suddenly having limitless operating hours and limitless labour makes you rethink what consumers and clinicians might want from your service.

Part of the process of using these agents is also examining what your human staff are currently doing. We are fortunate to have systems in place to review what types of calls we receive and what types of calls we could target. We decided we really wanted to replace those scripted calls.

Do you bulk bill? Are you taking new patients? When’s the pathology room open? What’s your fax number?

While appointment bookings form a part of the process it’s not actually the main target here. We already offer online bookings so replacing HotDoc/AutoMed/other was not really the aim.

We decided initially to run an inbound FAQ and new patient enquiries agent and have seen a clear drop in these calls coming to our human staff.

One of the surprising alternative use cases has been outbound calls (which is way more useful that I anticipated).

Traditionally I might offer patients a few options on how to proceed during a consult — get this investigation done if that doesn’t go away, or take this antibiotic if things don’t improve by day five.

We now have an agent call the patient to follow up with defined questions and goals. A summary is generated to the GP’s inbox, who can then choose how to proceed, advise or action.

This is not a headcount replacement. Instead we are seeing the tool create new opportunities to follow up on consumers and feed higher quality information back to the GP.

We’ve also found outbound agents are great for pre-consult data collation and we are shifting to a model of proactively offering appointments to discharged patients – a workflow where a GP can trigger an agent to follow up the patient and find a post-discharge appointment slot at the point of reviewing the discharge summary in the inbox.

Finally, there is no playbook in the broad and evolving field. It takes imagination to think about workflows and how a new technology can enhance the consumer experience.

While I can’t call 30 patients from today to see how they got on, next week I can review 30 summaries of agent calls and ensure those who need more attention drift to the top of my review pile.

My staff can spend more time speaking to those who need real human input when my agents field the questions about what Dr Max’s provider number is and whether they need to book for those fasting bloods.

I’m really excited to see how this field evolves and what creative use cases come from clinicians harnessing and sharpening their use of agents.

It’s an exciting time for the next iteration of primary care AI wars.

Dr Max Mollenkopf is a GP and practice owner in Newcastle, NSW. He receives no payments from Heidi Health and participated as a beta testing clinic for the voice AI scribe product.