AI Engineering Foundations with Claude

By the end of this lesson you will understand what a large language model actually is under the hood — tokens, context windows, temperature and sampling, and why models sometimes fabricate confident-sounding but wrong answers. It is the foundation for every course in the Claude AI Engineer path; later lessons build directly on the mental model you form here. No prior LLM or AI experience is assumed.

By the end of this lesson you will know your way around the Claude Console at platform.claude.com — where API keys are managed, where usage and cost are tracked, and where the Playground lets you test prompts interactively — and you will be able to pick the right Claude model tier for a given task from the current family (Haiku 4.5, Sonnet 5, Opus 5.5, Fable 5.1). This is the second teaching lesson in the Claude AI Engineer path; L3 immediately puts these choices into Python code.

By the end of this lesson you will know exactly how to shape a Claude request — where the system prompt goes (a top-level parameter, not a message with role="system"), how the messages array carries conversation state across turns, how to stream responses token-by-token, and what each stop_reason value tells you about how a call finished. This grounds the message shape you saw in L3 and prepares you for the multi-turn REPL in L5.

By the end of this lesson you will be able to compute the USD cost of any Claude request from token counts and per-model rates, understand how prompt caching and the newer tokenizer affect that cost, and factually position the Claude model family against GPT and Gemini for a working developer's decision. This is the last teaching lesson before the CLD-AI-101 capstone, where you turn model selection into a written recommendation backed by real benchmark numbers.