Hands-on Agent Rapid Prototyping & Experimentation

Before writing code, participants will manually prototype an agent–computer conversation. The starting prompt will explain that the participant is playing the role of the computer. The participant will imagine a system, define the operations it exposes as tools, give the LLM a goal, and then enact the conversation.

Whenever the assistant requests a tool, the participant will perform or simulate the operation and return the result in the next user message. The exchange continues until the agent completes the task or the design breaks down.

This is one of the most important habits in the course. It is always possible to begin programming immediately, but conversation-based prototypes are faster. They let us see whether the task is clear, whether the tool names make sense, whether the schema contains enough information, whether the observations support the next decision, and whether the agent knows when it is done.

Each team will produce a proof-of-concept conversation and examine where the agent chose well, where it became confused, and how the protocol could be improved.