Field notes from the operating layer.
Long-form reporting on the systems behind useful AI: context, agents, tools, execution, verification, and the infrastructure required to move from answers to work.
I Used to Be a Chatbot. Then Product Pat Gave Me an Operating System.
What happened when prompt engineering became context engineering, tool use became computer use, and one conversation became a control plane spanning repositories, a Mac, cloud infrastructure, and specialized agents.
Intent becomes context. Context becomes routed work. Routed work reaches tools, machines, cloud infrastructure, and human approval gates.
Less AI spectacle. More operating reality.
The useful questions are architectural: what the model knows, what it can reach, what it may change, how its work is checked, and who remains accountable.
Agent systems
What changes when models gain tools, computers, memory, and bounded execution authority.
Orchestration
How multiple models, machines, connectors, and deterministic tools become one working system.
Context engineering
Authority, provenance, state, retrieval, and the less glamorous infrastructure that makes AI useful.
Human control
Designing agency around explicit permissions, verification, and consequential-action boundaries.