Insights on technology, AI & leadership
First-person notes from the work — how I think about hard technology calls, where AI actually pays off, and building in a new country.
They thought they needed to hire. They needed one system.
A Canadian consumer-products company had hit a revenue ceiling and assumed the answer was a hiring round. The real constraint was not demand. It was that every order had to be entered by hand.
You can't give an agent 266 tools.
Scaling an agent's reach to hundreds of government-data tools turned out to be a problem of discovery, not enumeration — findability and context discipline, not a bigger model.
We didn't need a better model — we needed the project's context
A bought AI tool aced generic training content and fell apart on specialized pharma work. The fix wasn't a better model. It was a graph of the project's real context.
The double leap of faith: rebuilding from zero in a new country
First-person lessons from arriving in Canada and rebuilding from zero — twice — about networking as a second profession, the no you already have, and doing business with people.
You probably don't need AI — you need someone to codify the logic
Where AI actually earns its keep: embedded in your core process with a human in the loop, not bolted on as another chatbot.
Most technology decisions are judgment calls, not right answers
Why I treat the hard technology choices as judgment under uncertainty — and why my job is to help a team decide, then hand the decision back to them.