Foundations: design your first AI agent
Understand what an agent really is, and learn to delegate one real task effectively.
1. Using AI vs. directing a system
Most people chat with an AI one prompt at a time. Building an agent means giving a system a clear goal, the context it needs, the tools to act, and a boundary it must not cross — then letting it run a repeatable piece of work. Day 1 is about shifting from using AI to directing it.
2. The vocabulary that matters
- AI — the model that thinks and generates.
- Agent — an AI that is given a job, tools, and limits, and works toward a goal.
- Workflow — a repeatable sequence of steps an agent follows.
- Cron / schedule — a trigger that runs a task on a fixed clock.
- Event / webhook — a trigger that fires the moment something happens.
3. The agent's anatomy
Every useful agent has six parts:
- Goal — what success looks like.
- Context — background the agent needs.
- Reasoning — how it thinks about the task.
- Tools — what it can use to act.
- Action — what it actually does.
- Feedback — how you review and improve it.
4. The orchestrator mindset
You are not writing code. You are delegating — deciding what to hand off, to which agent, with which instructions, and where a human must still approve. The best starting point is a recurring task with a clear input and a clear output.
5. Common beginner mistakes
- Giving an agent a vague goal with no definition of done.
- Handing it too many tools or too much freedom at once.
- Skipping the approval step on anything commercial or external.
- Expecting one giant "do everything" agent instead of small, reviewable workflows.
6. A simple agent loop
An agent receives an input → reasons with its context → uses its tools → produces an output → is reviewed → learns from feedback → runs again. Day 1 explores this loop on paper with a real example: turning meeting notes into clear next actions.
Your assignment
- List five tasks in your business you could delegate.
- Pick one painful, recurring task with a clear input and output.
- Write down: the selected task, the outcome you want, the context needed, and the approval limits.