You go to the doctor, there's a little device on the counter or a sign on the wall, and you sign the form that says an AI will record the visit. You don't really think about where that recording goes.

Ross Mitchell thinks you should. He's the Alberta Health Services chair for AI in health at the University of Alberta, and he has two questions he wants every Canadian to ask their doctor. Where is my data stored, and is it being used to train the vendor's model? Your doctor probably can't answer either one. 

My colleague Jennifer Friesen wrote about why these questions matter beyond your doctor’s office.

If your organization is running customer data through an LLM from one of the big frontier model providers, you may not like your own answers any better than your doctor's.

Mitchell's team in Edmonton built their own AI scribe, called Jenkins, inside the Alberta Health Services firewall, when they could have bought one off the shelf. 

An AI scribe listens to a doctor's appointment and turns the conversation into clinical notes, the kind of paperwork that used to eat a physician's evening. 

Jenkins runs on Snowflake, a cloud platform companies use to store and work with their data, and, crucially, the patient audio never leaves the protected environment. It’s also estimated to cost under $30 per seat a month.

Jenkins uses GPT-5, with other models swappable behind it. Mitchell treats the frontier model as a part he can pull out and replace. 

The value is in everything around it, the firewall, the data controls, the benchmark his team built to define what a good clinical note looks like.

This is one of the arguments Benedict Evans makes in his latest presentation, AI eats the world. 

Evans, formerly a partner at the venture firm Andreessen Horowitz (a16z) and now an independent analyst, is a closely followed figure in the tech industry. He argues that AI is the biggest change since the internet and the smartphone, but he’s not convinced it’s necessarily bigger.  

He believes that frontier models may slide toward commoditization, causing value to move up the stack into the systems, data, and workflows built on them. Frontier LLMs will become the telco-like infrastructure upon which others build innovation. 

Mitchell's work is an early example of this in the wild. 

While many organizations argue over which model to license, his team treated that question as settled and went to work on the part that's theirs to own.

Some light reading for when the vendor demo runs long

Canada puts $100M behind national health data network. Jennifer Kervin reports on VITAL, a platform linking de-identified records from 160 hospitals to give health AI a national foundation. 

Three in four large enterprises have rolled back AI agents. Customer data getting exposed was the top reason. Probably cheaper to ask the hard questions before launch than to unwind it later.

The Chief AI Officer rush is an ownership question. Three in four organizations now have a Chief AI Officer, up from a quarter a year ago. It's unclear whether a new title clarifies who owns AI or just scatters the accountability further.

Your plan for the next system outage may be built on wishful thinking. Curtis Simpson told this year's CIOCAN Peer Forum that passing the audit isn't the same as surviving a real outage. He figures 95% of organizations have never tested recovery the way an actual failure would.

Cycling industry bets on smart bikes to boost sales. After three straight years of falling sales in Europe, the industry is betting AI can help, with motors that read your heart rate and systems that warn you about traffic. 

Speaking of riding bikes for a good cause. In a few weeks, I’ll be riding the Princess Margaret Cancer Foundation's Ride North. Our family has raised more than $53,000 over the last four years and is looking to add to that number this year.  My 13-year-old daughter is kicking my ass (again) with another $12,000 so far this year. I'm proud and a little embarrassed she's so much better at this. Your donation helps me catch up, and helps Princess Margaret do the part that matters.

A few years ago I was at a restaurant in Waterloo, back when Stephen Hawking was a Distinguished Research Chair at the University of Waterloo’s Perimeter Institute for Theoretical Physics. 

He was at the same restaurant one night, and his voice is not one you miss in a crowd. The thing I heard him say most that night was "I don't know." I found it telling that one of the great minds of his era said it so easily. There's no shame in not knowing.

Evans makes a point of saying the same thing. It's early days in mainstream adoption of AI. He could easily be wrong about any of his predictions, and nobody knows for sure. 

His advice is to presume radical uncertainty. That doesn't mean standing still. It means putting your attention on the things you can control while the rest stays unsettled.

— David