STARTING A RUN

LEARNING GOALS You will learn what happens in the first moments of an agent run, before the model has even been called once. You will learn about task intake, which means receiving and structuring the user's request. You will learn how instructions are assembled and goals are defined. And you will learn how state, budgets, and permissions are initialized, and why the order of these steps matters for safety. WHAT IS A RUN A run is one complete attempt by an agent to accomplish one task, from the moment the request arrives to the moment the agent stops. A single run may involve many loop iterations, many model calls, and many tool calls, as you learned in Part Three. Some people use the word session for a longer conversation that may include several runs, and the word turn for a single exchange within a run. Because a run is a unit of work, it is also a unit of accounting. We track the cost of a run, the time of a run, the outcome of a run, and the trace of a run. For all of that to be possible, the run needs a clean, well-defined beginning.