Product notes and technical essays on durable agent applications.
Part 5 turns the measurement series into a practical decision: start from the worker workload, then choose the harness profile that fits it.
Part 4 looks at the costs that show up after launch: memory per active worker session and total tokens per completed task.
Part 3 compares config-time tool composition with runtime tool loading so teams can match a base harness to the shape of the worker's tool catalog.
Before token cost matters, a base harness has to support the job your worker must do. Part 2 checks MCP, skills, hooks, subagents, code execution, shell access, and web access across five candidates.
Teams building worker agents need to choose a base harness for harness engineering. Part 1 measures how much context each candidate carries before the worker starts work.