Writing
Essays on agent architecture — what the harness actually does, why composing patterns matters more than picking models, why most agent failures are budget-allocation failures in disguise. Originally published on Substack; kept here as canonical archive.
-
May 12
Why Every Agent Team I Know Runs Out of Context Before They Run Out of Compute
Different teams, different domains, different stacks. They keep hitting the same wall well before they hit any compute or model capability limit. Context is the bottleneck, and unlike compute, you cannot scale it horizontally. The model spends; the harness budgets.
-
May 1
"Build Me an Agent" — The Most Dangerous Sentence in Software
Four months and a quarter-million dollars in compute later, the multi-agent system was stripped back to a single ReAct loop with three tools. Latency dropped 7×, cost down 90%. It was not a bad build — it was a build for the wrong problem.
-
April 6
The Five Chaos Problems in Agent Design
We are in 2026, and 'agent' has become the most overloaded word in software since 'cloud.' Five specific chaos problems the field has not solved — and the structural framework that motivated me to write a book about it.
-
March 30
What 19 Papers Taught Me About Where Agent Engineering Is Really Going
I pulled every significant agent paper, blog, and doc from the past three years — Anthropic, OpenAI, Berkeley, Simon Willison, Lilian Weng, Eugene Yan, and a dozen more. Nineteen sources later, the field had converged in directions most practitioners had not yet noticed.