About
Working Notes on Agent Systems publishes field notes and proofs from the agent engineering frontier: harness engineering, long-running workflows, retrieval, AI-native workbenches, and technical distribution.
Brad Zhang writes and publishes this site. His public work circles one question: what has to exist around a model before an AI workflow becomes reliable software. The answers usually land on harnesses, memory, evaluation, and workbenches.
Biography functions here as authorship and accountability, not as the main character. The site serves the arguments first and the author second.
Collect
Pull public work back from platforms.
X field notes, longform essays, and GitHub repositories are treated as one proof system instead of three disconnected channels.
Reframe
Sort by question, not by feed mechanics.
The site reorganizes content into recurring engineering questions so readers enter through judgment and proof rather than timelines alone.
Compound
Turn writing into durable public infrastructure.
Projects, essays, and notes are built to reinforce one another rather than compete for attention.