Part 1 · The idea · 🟢 Easy · ~25 min

What you learn in this lesson: what an agent actually is, the five words around it — model, skill, harness, knowledge base — and the one rule of delegation that makes all of them work.

🎬 Watch it first (1¼ minutes)

🥂 Day one at Rosehill

You have just taken over Rosehill Catering — a two-person studio, you and chef Lena. On your desk: a stack of order notes, a price list, and a phone that keeps ringing.

And one more thing. As of today you have a new coworker. It has read every business book ever written, types faster than you can think, never sleeps, never gets bored — and knows absolutely nothing about Rosehill. Not your prices. Not your clients. Not that the kitchen is closed on Mondays.

That coworker is an AI agent, and this lesson is your first day working together.

📖 What an agent is — and is not

You have probably used a chatbot: you ask, it answers, you copy-paste the answer somewhere. An agent is the next thing: software that does rather than merely answers.

The difference is not intelligence — it is hands. An agent can read your documents, search your workspace, and produce work you review. Which is why working with one feels less like using software and more like managing a colleague: you brief, it works, you check, you decide.

🗣️ The five words

Every trade has its handful of words. These five are the whole vocabulary of this course — each with the business picture that makes it obvious.

1 · Model — the brain you hire. Models come in sizes, like seniority levels. A small model is the quick junior: fast, cheap, fine for routine work. A big model is the senior specialist: slower, pricier, worth it for the tricky contract clause. You don't hire a senior for every errand — you pick the brain per task.

2 · Agent — the coworker. An agent is a model plus a job description (its instructions), plus tools (what it may read and use), plus permissions (what it may touch). Same brain, different job description → a completely different colleague. This course ships three: Advisor explains and checks your work; Planner drafts and structures on instruction; and Coder — you meet it in lesson 2 — builds, on the strongest coding model your deployment has, because a wrong build costs more than a slow one.

3 · Skill — the recipe card. Work you do the same way every time deserves a written procedure — in the agentic world that procedure is a skill, and you trigger it by typing / and its name in the chat. Think of the laminated recipe cards in Lena's kitchen: anyone can execute one, nobody has to remember the steps. And here is the part most people don't expect: a skill can have a UI part — before it runs, it can pop up a question on the screen ("Who should take this on?") and use your answer. You will click through two of those in lesson 2.

4 · Harness — the workbench. Where does the agent actually sit while it works? The harness is its workbench: the chat window, the tools within reach, the safety rails around it. The same model can work at different workbenches — a chat here, a coding terminal elsewhere — the way the same chef can cook in different kitchens. You rarely need to change the harness; you just need the word, so the day someone says "that runs in a different harness" you hear "different workbench", not magic.

5 · Knowledge base — the company handbook. This is the big one. The agent's general knowledge is enormous; its knowledge of your business is exactly one thing: what is written in your knowledge base — here, the Rosehill Handbook: company facts, menus and prices, house rules, clients.

The golden rule: an agent knows nothing about your business except what is written in the knowledge base — and what you tell it in the brief. Change the handbook, and you have changed what your agent knows. That is not a limitation. That is the steering wheel.

🧭 When to use what

Five words, one decision each. This little map is the lesson:

⚡ See it — the golden rule, live

Try it right here — this is a real composer, and sending opens a real conversation:

Ask two questions, in order:

1. “What does the Harvest Table cost per guest?” (the ✨ starter) — the Advisor answers CHF 72, instantly. That fact is written in the handbook, so the agent knows it.
2. “What did we charge Meier & Co last year?” — it tells you, honestly, that this is written nowhere it can read — and refuses to guess.

Same agent, same brain, seconds apart. The difference between the two answers is not intelligence — it is what is written down. Remember the feeling of answer 2: in lesson 3 you will make exactly that kind of question answerable, by writing the fact into your own handbook.

💬 Where the chat lives. You never need a special page for it: the chat is always one click away in the top menu, on every page of the portal — click here and watch the menu show you exactly which button. A conversation you start there automatically carries the page you are on as its context, so the agent knows where you are standing. Lesson 2 will use that fact deliberately.

✍️ Your turn

Three tasks — each a puzzle that sharpens your judgement, then a real conversation with a course agent.

• 🟢 Easy · Meet the words — seven situations, five words; then the Advisor quizzes you on new ones.
• 🟡 Medium · Your first brief — spot what will go wrong in four briefs, then fix one and send it for a real quote.
• 🔴 Hard · The right tool — six situations, six tempting wrong moves — and one trap you then play for real.

Start with the easy one. The brief you fix in the medium task gets sent for real — and the quote that comes back is graded by you.

❓ Check yourself

Six quick questions — with instant answers.

Course overview: Agentic Business Primer · Next: Lesson 2 — Wish for a tool »