Writing
Notes from the build.
Essays and field notes on LLM systems, retrieval, agent architectures, and the unglamorous engineering that turns a demo into a product.
Aug 28
2026
Stop Spoonfeeding the Model
The fastest way to make an agent worse is to explain too much. On over-instruction, prompt debt, and briefing agents instead of scripting them.
Jul 21
2026
What an Agent Tells Its Future Self
Long-running agents need to hand off to their own next run. On the split between history, state, and knowledge — and the smallest checkpoint-and-resume loop that actually works.
Nothing here yet
No posts match that filter.