How to use this reference
- Work through topics top to bottom — MCP and agents assume you understand tool use; RAG assumes you understand embeddings and hallucination limits from Topic 1.
- Every subtopic teaches what the model is doing so you can verify outputs, set guardrails, and choose the right integration pattern — not copy-paste prompts blindly.
- Vendor docs (Anthropic, OpenAI, Cursor) change fast — our overviews point to durable concepts; follow the linked official docs for API details.
- Optional tasks include local MCP experiments and workflow design drills — skip what you already do daily at work.
- Building production AI products at scale (serving, evals, rollouts) is a separate track — continue to AI Systems Reference after this one.
The Reference
- 1
Transformers, tokens, and inference — what the model actually does before you prompt it.
- 2
Prompts are API calls — context is the payload. Design both like you design interfaces.
- 3
IDE agents, Cursor workflows, and the daily loop — where AI earns its place in your toolchain.
- 4
Ground the model in your docs and code — retrieval beats bigger context windows.
- 5
Give the model hands — tool use, MCP servers, and CI integrations that turn chat into action.
- 6
Single agents hit ceilings — orchestration, handoffs, and evals for production agent systems.
- 7
Route the right model to the right task — and know when fine-tuning beats RAG beats prompting.
- 8
Scale AI adoption across a team — rules, skills, security, IP, and policies that outlast any model version.