Blog
Four years of coding with AI
In 2022 every line of code I shipped went through my own hands. By 2026 most of it is written by agents on machines I rarely touch. One diagram, one change at a time.
On agent swarms
Human attention limits how many agents a team can run. Scaling further requires agents that supervise other agents and software they can operate themselves.
How will we ask AGI?
As execution becomes cheap, deciding what deserves doing becomes harder. Delegation, judgment, and written authority will shape how we ask capable systems to work.
Agentic Harness Archaeology
Agent harnesses preserve workarounds for older models. A look through RAG, tool-call parsers, workflow graphs, and sprawling instructions to decide what still earns its place.
On local inference
The unglamorous ongoing labor of keeping a private local inference rig running after the romance of self-hosting has worn thin.
Hyperscalers are 4500 years old
The Great Pyramid and a hyperscale datacenter are structurally identical bets — concentrated short-term cost justified across a longer horizon.
A History of Local LLMs
A detailed look at the timeline of local large language models.