Eleven modules from the language-model basics to a shipped, production capstone. Built for small-business owners and operators — assumes zero prior AI experience. Interactive playgrounds, quizzes, and three capstone tracks you can pick from. Real examples across agencies, retail, real estate, professional services, and content businesses. No hype.
Five quick questions and I'll route you to the right module. Nothing is graded and you can always take the long path.
Modules 1–4 are concepts, 5–7 are patterns, 8–10 are hands-on builds, 11 is the capstone project. Mark modules complete as you go — the tracker follows you across sessions.
Language models, tokens, context, temperature, and why 'talking to AI' works. Set the mental model you'll use for the rest of the course.
Structure, role, examples, guardrails. The prompt patterns that outperform 'be more specific' every time.
How a language model becomes an app. Function calling, browser use, connectors — and where Perplexity Computer fits in.
Workflows vs agents. The six canonical agent patterns from Anthropic, illustrated with agency examples you can run tomorrow.
When determinism beats intelligence. Build an inbound-lead intake flow in n8n and a Cloudflare Worker version alongside it.
Retrieval-augmented generation for your business. Embed your past proposals, quotes, or docs; get grounded, cited answers.
The Model Context Protocol lets any agent plug into any tool. Install one, use it in Claude / Computer / Cursor.
Skills, subagents, cron, connectors. Stand up a weekly lead-research agent that runs while you sleep.
Copy-paste JavaScript for a working agent that uses tools, retries on failure, and returns structured JSON.
Costs, evals, guardrails, PII, and the questions to ask before any AI touches a client. A pragmatic ship-checklist.
Three ship-ready tracks: Proposal Assistant, Content Ops Engine, or Support Copilot. Pick the one that saves you the most time, then build the smallest useful version this week.
Every module starts with plain-language explanation and a diagram. No jargon without definitions — hover any underlined term to see it defined.
Modules 2 and 9 have live playgrounds. Test prompts against real APIs or watch an agent's tool-calling loop run in real time.
Each module ends with 3–5 questions. Pass with 70% and the module marks itself complete. Retry anytime.
Every module ships with a downloadable starter file — an actual prompt, workflow, or code file you can use tomorrow.
The vocabulary of AI, in plain language. Every term here is also hoverable inline inside the modules — no need to memorize.
| Term | Meaning |
|---|---|
| LLM | Large language model. The prediction engine at the core of every AI product. |
| Token | Roughly ¾ of a word. Models charge per token. |
| Context window | How many tokens the model can hold at once. Its short-term memory. |
| Prompt | The input text you send. Includes system instructions + user message + prior conversation. |
| System prompt | The always-on instruction at the top of every conversation. Sets role, tone, rules. |
| Tool / function calling | Letting the model call a function you defined (weather, search, send-email). |
| Connector | A pre-built package of tools for a popular app (Gmail, Notion, Slack). |
| Workflow | A predefined sequence of steps involving LLMs. You draw the flowchart. |
| Agent | A system where the model picks its own steps and tools to reach a goal. |
| RAG | Retrieval-augmented generation. Look up relevant chunks first, then ask the model. |
| Embedding | A vector that represents the meaning of a chunk of text. |
| Vector database | A DB optimized for finding similar embeddings quickly. |
| MCP | Model Context Protocol. Open standard so any tool plugs into any AI app. |
| Eval | A test suite of inputs + expected outputs to measure model quality. |
| Guardrail | A rule that prevents an agent from doing something dangerous. |