All posts

What a 2.4B Vision Model Can Actually Do on an iPhone, Offline
Ten days testing whether a small open-weights vision model can do real field-service work on a phone with no signal
Aug 2026
The Canadian Roots of the AI Boom
From Hinton's lab at U of T to AlexNet, the transformer paper, and Cohere coming home to Toronto.
Aug 2026
Stop Blaming the Model: You Have a Context Problem
The eight layers of context I bring to every phase of product development, and why the real win is building them into how your team works.
Aug 2026
Your PMs Shouldn't Be Making Micro-Product Decisions Anymore
As coding loops compress, engineers have room to own the product calls inside a feature. That scales PMs rather than diminishing them.
Aug 2026
The Circularity of Error: Why Agentic Dev Needs a Verification Loop
A green test suite only proves the code agrees with the tests. Here's the four-part framework we built to break the self-consistency trap.
Jul 2026
When Your AI Writes Code, Tests Become Your Seatbelt
Adding snapshot tests, service tests, and Maestro E2E flows to a small iOS app in one afternoon, and writing the documentation an AI maintainer actually reads.
Jun 2026
Wiring Cohere's Command A+ into Apple's Foundation Models Framework
A first-experience writeup on making Command A+ a drop-in replacement for Apple's SystemLanguageModel
Jun 2026
Canada's AI for All Is a Smart Strategy
Why the sovereignty piece of Canada's national AI strategy is the part that matters most
Jun 2026
Shipping Faster in the Age of AI, Without Breaking Trust
The problem was never shipping too often. It was shipping disruption too often. A change taxonomy, a soak lifecycle, and giving users a way to opt out.
May 2026
The PRD Is the Highest-Leverage Artifact in Software
Why great PRDs are finally within reach, why the prototype is not their replacement, and why both points sharpen when an agent is the reader.
May 2026
Five Things I'm Seeing on Teams Where Claude Code Is Actually Working
Some teams are seeing 2x on throughput and some are seeing very little. Five common threads separate the two cohorts.
May 2026
The Bottleneck Was Never Writing Code
How small PRs and human-centered review become more important, not less, in an AI-assisted world
Dec 2025
How We Scaled AI Across Our Mobile Engineering Team
We rolled out AI coding tools for writing code. Seven months later, engineers use them just as much for design docs, test plans, and PRD reviews.
Dec 2025
Ship the Simplest Thing
Most enterprise teams are optimizing for imaginary futures instead of real presents.
Dec 2025