Module I — From Model to Agent System

Status: outline. Lecture body not authored.

The object of study is the system that controls the agents. Holding the model fixed and changing the harness can change performance materially. The result depends on Model × Harness, not on either one in isolation.

Model ≠ Agent

• probabilistic model
• agent loop
• environment
• tools
• context
• execution
• state

AGENT SYSTEM  =  MODEL  +  HARNESS  +  ENVIRONMENT

Unique process — Harness Swap Experiment

Hold:

MODEL = CONSTANT
TASK  = CONSTANT

Vary only HARNESS. Measure success rate, tool calls, tokens, latency, failures.

Lab: ../labs/L1-harness-swap.md