// notes
Working notes on AI operations.
The thinking behind the work. Written when there's something worth saying.
Software used to be metal. Now it’s clay.
SaaS forced everyone to bend their operations around someone else’s workflow. The collapse of code cost changes the equation: businesses can own sovereign, custom operational infrastructure shaped around how they actually run — and stop looking like the market.
full essay in progress
A mech suit for knowledge work
The four levels of AI in an organization: structured process over clean data, the powered-up individual, the org where agents have their own handles, and the management layer everyone skips — grading, training, and failure. An agent that isn’t graded drifts.
full essay in progress
Never trust a single pass
The principles that keep autonomous systems honest: ground everything or don’t answer, a human gate before anything expensive or irreversible, defense in depth, observability as a feature. These are the controls that stop confident fiction.
full essay in progress
The always-on shop
Field report: running a firm where the AI works nights. What the overnight loop actually does, what the morning approval looks like, and what breaks when you skip the guardrails.
full essay in progress