Learn to Build Agent Harnesses

“Agency comes from the model. An agent product = model + harness.”

“One loop & Bash is all you need”

“Adding a tool means adding one handler”

“An agent without a plan drifts”

“Break big tasks down; each subtask gets a clean context”

“Load knowledge when you need it, not upfront”

“Context will fill up; you need a way to make room”

“Safety is a pipeline, not a boolean”

“The loop owns control flow; hooks observe at named moments”

“Only persist what cannot be re-derived from current work”

“The model sees a constructed input pipeline, not one giant static string”

“Most failures are signals to try a different path”

“Break big goals into small tasks, order them, persist to disk”

“Run slow operations in the background; the agent keeps thinking”

“Time itself becomes another launch trigger”

“When the task is too big for one, delegate to teammates”

“Teammates need shared communication rules”

“Teammates scan the board and claim tasks themselves”

“Each works in its own directory, no interference”

“External capabilities join the same routing as native tools”