Run your agents on the hardware you already wrote off
Agents run on old, fully managed machines. Per machine the electricity costs about as much as a coffee a month. Comparable AWS capacity costs $240 to $475.
My blog posts.
Practical notes on applied AI, engineering leadership and building systems that hold up beyond the demo.
You cannot automate what you never wrote down
AMOS is the orchestration layer I said I would build in 2026. Nine months in, the constraint is not agents, models or protocols. It is that the processes we want to hand over have never been written down in a form anything could follow.
A skill is not a markdown file
Everyone is writing agent skills right now, and most of them quietly make the agent worse. What that actually costs you, why "be thorough" is the wrong instinct, and the checks that have been gating every skill across an engineering organisation for months.
No AI project. Just a new hire.
What twenty teams taught me about treating an AI agent as a colleague with a defined role, identity and responsibilities.
I'm not building an AI Agent. I'm hiring a virtual employee.
Everyone's talking about AI agents. I keep coming back to a different question — in a remote-first world, what actually separates a human colleague from an AI one?
2025: The year AI learned to work (and I still didn't write enough blog posts)
A reflection on how AI agents transformed software development in 2025, my goals for building AMOS in 2026, and why powerful tools still require human discipline.