Hi {{first_name}} ,

Most technology leaders aren't going to sit in a meeting and tell their CEO the board has no idea what it's asking for. A confident few might, but many won't, fearing it would surely be a resume-generating event.

But Jim Balsillie said as much this week, out loud, to a room of directors. General business judgment, he told them, can't run a company built on AI, IP and data.

It was at the 2026 Innovation Governance Summit, put on by the Council of Canadian Innovators in Toronto, and my colleague Jennifer Kervin sat in to cover the event.

At the summit, advisor and board member Carol Leaman said what most technology leaders already know and can't say out loud.

"You guys are trying to play in a baseball game. The game hasn't even started for you yet. Other companies are in the sixth inning, and you're sitting on the benches waiting for the game to start," said Leaman.

Here's what that means for you: the board that can't tell a data strategy from a data centre is the same board approving, or killing, your AI budget (likely right after someone's grandson showed them ChatGPT on their phone at the lake). So when you bring them something that matters, like data control, vendor lock-in, or a model your whole business runs on, it can get killed simply because nobody at the table understands what you're asking for.

The people signing off on AI spend should understand what they're buying, who controls it, and what happens if it goes away, because when they can't answer that, the risk isn't the AI, it's the board.

→ So if you want to tell your CEO that a generalist board is going to tank company value, or you're the CEO who needs to tell the board, send them this: The era of the generalist board director is over.

The trouble with renting AI by the token is that the meter never stops, and the companies running it hardest are the first to feel it. I went through Mozilla's first open-source AI report, which found Uber burned through its annual AI coding budget in four months while Stripe moved onto open models it runs itself and cut its bill 73%. Somewhere an Uber finance exec aged a decade watching the coding team expense a small car every week.

What makes that move possible is that open models have nearly caught the closed ones on everyday tasks, while the cost of running them has fallen roughly 50 times in three years. Closed models still lead on the hardest reasoning and coding, and Anthropic's Fable 5, which landed after the report, is a beast at both.

What makes leaving hard is something Mozilla calls "the harness," a term many leaders may not have come across yet but will need to understand. If the model is the engine, the harness is everything that turns it into a working vehicle, the software that gives the AI a memory, connects it to your tools, and decides what it's allowed to do on your behalf. Each vendor now tunes its harness to run best with its own model, so switching means replacing the tool your team works in every day, not just the model underneath.

It's the gym membership model, easy to join, designed to make leaving a chore. That's where the lock-in lives now, and it tightens the longer you stay with it.

Owning isn't free either. It takes people and hardware most mid-market shops don't have, so the point isn't to run everything yourself. It's to decide what you own and what you rent on purpose, before the renewal decides for you.

If you needed proof the risk is real, Anthropic pulled global access to its Fable 5 model for 19 days last month to meet U.S. export controls. The only people having a good week then were the lawyers, who got to explain "force majeure" to the tech teams who swore they'd read the contract.

Know whose stack you're on, and keep a second option warm enough to switch to.

→ Dive into the findings: What tech leaders should take from Mozilla's inaugural open source AI report

Toronto's Vector Institute has released a free, open-source tool called UnBias-Plus that flags biased language across race, gender, age and political framing, explains why it flagged it, and rewrites it.

A line that reads "noncompliant patient refused medication" becomes "patient declined medication," and that one word stops trailing someone into every appointment after.

"The people most harmed by biased language are often the last to know it's there," says Shaina Raza, applied machine learning scientist for responsible AI at Vector.

Bias in one document hurts one person, but bias in the data a model learns from gets repeated everywhere the model runs. If your team fine-tunes on your own data, this is the cheapest way to catch it before it's baked in, and it's the kind of thing your board will eventually ask whether you did.

→ Read more: A new Vector tool screens AI training data for bias

While the boards workshop their AI vocabulary, some Canadians got on with it.

My colleague Jennifer Friesen shared a story we heard at Upper Bound, one of Canada's biggest AI conferences, where Alberta's deputy minister of technology and innovation, Janak Alford, showed how the province pointed roughly 50 AI agents at its own code. He had built it himself, and answered questions in real detail because it was his own work, not a vendor's demo.

The agents scanned 466 million lines in about 20 hours, work Anthropic estimates would take one person 6.5 years, and turned up a system of 1,280 applications doing about 700 jobs, including 27 ministries that each built their own version of the same login screen, apparently without telling each other. You can't fix duplication you can't see, and most organizations have no idea how much of it they're paying to run.

"I have not found myself personally working less," says Alford, on where the saved time went.

Scotiabank, Sun Life and TELUS have teamed with Lightworks to build shared AI systems while each keeps its own IP, a build-together, own-separately arrangement that already handles more than two trillion tokens a month.

A bank would never hand its core data to an outside vendor, and these three have decided the AI built on that data is no different — it’s too important to rent.

Vancouver's Conexiom is betting on the least glamorous use of AI there is. Instead of chatbots and pilots, it's automating the dull back-office work of processing orders and invoices, through an expanded deal with software maker Epicor, our contributor Dr. Tim Sandle reports. Their wager is that the money is in the boring jobs, not the flashy ones.

Two of these are ownership plays and one is about making AI earn its keep, but all three share a refusal to wait for permission. Before you spend a dollar modernizing, do what Alberta did and find out what you already run.

→ Read more: What Alberta found when it pointed 50 agents at its own code

Some light reading for when you've had enough of boardrooms and billing.

AI helps pathologists spot prostate cancer faster.
Read this one now if you run anything in health. A landmark UK study found the gains came when AI worked alongside experienced pathologists, and the piece lays out what Canada's system can borrow from it.

Your next AI prompt comes with an energy bill.
This one can wait for the weekend. The cloud is a warehouse of computers burning power and water, and Canada's cool climate and clean grid make it one of the better places to run AI.

Canadians want AI for the small money, not the big money.
Worth a skim if you sell to consumers. A TD survey found people are happy to let AI check balances and track spending, but 71% still trust a human over AI on retirement, mortgages and planning, and only a third would take AI's word over their parents'.

Three in four large enterprises have rolled back AI agents.
One of our top-reads from June, so we’re sharing it again. A Sinch survey of more than 2,500 decision-makers found 74% have pulled a customer-facing AI agent back after launch, and the rate rises to 81% at the companies with the best governance, because they're the ones who can see what broke. Customer data exposure was the top reason.

The thread this week is that renting AI is cheap while you're small and expensive once it's running everything. The free trial, the low token price, the pilot nobody meters all look generous right up until the tool becomes the way your business works, and by then the price is the vendor's to set.

The companies that came out ahead this week refuse to lose track of what they own. Alberta audited its own systems and found how much work it was doing twice without knowing. Scotiabank, Sun Life and TELUS built AI together but each kept its own IP rather than hand it to a vendor. One story is about seeing your own operation clearly, the other about holding onto control.

Most companies learn the cost of that dependence the hard way, when the bill jumps or the terms change. The ones in this issue are trying to know the answer before the question gets expensive.

— Chris Hogg

If you have any questions, you can also reply to this email and I read everything.

Do you have a story we should chase, or a tip you'd share with a peer?
Contact us here.

Know someone who should be reading this. Forward it on or
they can sign up here.

This issue is presented by Calgary Economic Development.
If you want your brand in front of the people running technology across Canada, you can sponsor a future issue.