Every time an agent gives you weak work, there is a reason inside the system. The useful skill is knowing where to look.
A practical way to understand an agent is a model operating inside a harness. The harness gives the model instructions, context, files, tools, skills, memory, permissions, time, and ways to check its work. Together, these parts determine what the agent can do and the quality of what it produces.
In this session, I will take an agent apart piece by piece using products you already know, including ChatGPT, Claude Code, Cowork, and Notion AI. You will see how the parts work together and learn a repeatable way to improve the output.
