"Agent" gets used as an autonomy descriptor, a product feature name, and a platform primitive, often in the same conversation. A field guide to the three meanings, using GitHub Copilot and Azure AI Foundry as the running example.
A live workshop experiment comparing a generic Copilot chat to a custom Backlog Generator agent on the same requirements-writing task, and why the gap is an argument for investing engineering rigor in designing your AI assets.
Why the fear and the excitement about AI aren't two different camps of people. They're the same person, on the same night, and both feelings are trying to tell you something useful.
A confession about half-finished side projects, why loving to code was never enough, and how AI tools finally turned that love into a habit of shipping.
The old consulting model of knowing more than your client doesn't survive contact with AI. Here's what's actually happening, and what the new model looks like.
Two years into working with AI tools every day, here's my honest assessment of what's changed, what hasn't, and why I'm cautiously optimistic about the future of software engineering as a profession.
The previous post closed with a claim: get the artifact chain right at the team level and you have something replicable. That’s what program-scale AI adoption actually looks like.
Getting AI to produce consistent outcomes across a large program is hard. Most teams are discovering this the wrong way: each developer finding their own workflow, context evaporating between sessions, features drifting from intent, and no two teams using AI the same way.
In my last post, I wrote about how senior developers are getting disproportionate productivity gains from AI-assisted development. Multiple 2026 studies confirm the pattern: the more experienced a developer is, the greater the productivity impact. A Fastly survey put a specific number on it — senior engineers generating AI-assisted code at 2.5 times the rate of junior developers.