Blog
On agent swarms
The bottleneck is decisions per hour. Scaling past one human per agent means agents supervising agents, and a control surface built for machine operators.
How will we ask AGI?
When execution is close to free, the bottleneck moves to knowing what is worth executing. How delegation, taste, and paperwork reshape the way we talk to capable systems.
Agentic Harness Archaeology
Most of an agentic harness is fossilized workaround — dating the layers of RAG, tool-call parsers, DAG runners, and giant instruction files, and deciding what deserves to survive.
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.