The module covers the distinction between using AI to build software and building software that contains AI. Most of the course covers the first. This module covers the second: products where language-model intelligence is part of the user-facing value rather than only the development process.

Architectural questions arise when intelligence is a product component: where does inference happen, what latency is acceptable, how is the model prompted and by what, what happens when the model is wrong and who handles it.

Agent economics is treated with specific numbers because cost is a real architectural constraint. The module covers token pricing across providers, the impact of prompt caching, and the decision framework for frontier models versus smaller local models. The latency budget, defined as the response time acceptable for each kind of user interaction, determines the model choice more than capability does in most production systems.