Writing
Notes from the bench.
Lessons learned, observations and occasional essays about infrastructure, agentic systems, and the craft of making hard things understandable.
2026
-
The Points Behind the Points
What I said in the SNIA harness webinar, and the load-bearing ideas underneath: the dead corner, the menu and the transcript, visibility failures, build time versus run time, and autonomous action classes.
-
When LLMs Should Plan, Not Execute
Four reasons LLMs don't belong on the critical execution path, and the planner-actor architecture that can be used in production today.
-
Inference Illustrated: a build log
Notes from building an interactive curriculum for storage and networking people learning LLM inference. What went in, what came out, and the rules I would not break.