SESSION 1 · ~45M
The modern AI stack
Model, API, chat app, agent, IDE — these are layers of a stack, and confusing them is the most common beginner mistake.
People say 'the model' when they mean a chat app, and 'an agent' when they mean a prompt. The modern AI stack has distinct layers, and using them well starts with naming them correctly.
The five layers
- Model — the trained weights themselves (GPT-4-class, Claude, Gemini, GLM, Llama).
- API — the network interface that lets your code send prompts to a model and get tokens back.
- Chat app — a consumer product (the web chat) wrapping the API with memory, formatting, and safety rails.
- Agent — a system that lets the model call tools and run loops (read files, run code, browse) toward a goal.
- IDE — a coding environment (like Claude Code) that wraps an agent with file access, terminals, and project context.
Where Claude, Gemini, GPT, and GLM sit: each is primarily a model family exposed through its maker's API and chat app. Claude is also the engine inside Claude Code (an IDE/agent); Gemini powers NotebookLM and Google's products. Knowing which layer you're touching explains why capabilities differ across surfaces.
Open vs closed weights
Closed-weight models (Claude, GPT, Gemini) keep their parameters secret; you access them only through paid APIs. Open-weight models (Llama, GLM, Mistral, Qwen) publish the weights, so you can run them locally, fine-tune them, or inspect them. Open weights trade frontier capability for control, privacy, and zero per-token cost.
Pick the layer for the job. Need a quick answer? A chat app. Need it inside a spreadsheet or a script? The API. Need it to act across files toward a goal? An agent or IDE. Most frustration comes from using the wrong layer — trying to make a chat app do agent work, or hand-coding what an IDE already does.
TRY IT
List three AI things you used this week. For each, name the layer: model, API, chat app, agent, or IDE. Notice how often 'the model' was really an app or an agent — that reframing is the whole point of this module.
CHECK YOUR UNDERSTANDING
What is the key trade-off between closed-weight and open-weight models?
What does the module identify as the most common beginner mistake when working with AI?
Why does the module advise picking the right layer for the job?
NOTEBOOKLM
Connect your notebook
Paste a NotebookLM share link to embed your notebook for this module.