You'll leave with: a one-page “agent loop” for a task you really do, and your first copy-paste prompt.
Chatbot vs agent: the intern test
A chatbot answers. You ask “what should I cook?”, it replies, and the conversation waits for you.
An agent works toward a goal. You say “plan a dinner for four under $40 and give me a grocery list I can check”, and it breaks that into steps, does them, checks the result, and keeps going until the job is done (or it gets stuck and tells you).
At HumAIn we say an agent is a robot intern: fast, tireless, eager to help, and perfectly capable of confidently doing something slightly wrong. You wouldn't hand a new intern your credit card on day one. You'd give them a clear task, check their work, and widen their trust over time. Same rules here.
The agent loop
Almost every agent, from a simple prompt to a fancy product, repeats the same four moves:
- Goal: what “done” looks like.
- Plan: the steps it intends to take.
- Act: do a step (write something, search, calculate, fill in a spreadsheet).
- Check: compare the result with the goal. If it isn't there yet, go back to Plan.
What makes an agent more than a prompt?
| Piece | In plain words | Example |
|---|---|---|
| Instructions | The standing rules for how it works | “Always show your maths. Label estimates.” |
| Tools | Things it can do besides talk | Search the web, read a file, add a spreadsheet row |
| Context | What it knows right now | Your budget, your diet, last week's plan |
| Limits | What it must not do, or must ask first | “Never buy anything. Ask before sending.” |
Many “agents” you will meet are really a well-written set of instructions plus a few tools. That's good news: you can build the instructions part today, for free, by writing a prompt.
Your first agent-style prompt
Copy this into any chat. It gives a plain chatbot an agent's habits: plan first, show the work, check at the end.
Act as a careful planning assistant. My goal: [DESCRIBE THE GOAL, e.g. "plan a dinner for 4 under $40"]. Work in this loop: 1. PLAN: list the steps you will take (max 5). 2. ACT: do the steps and show your work. 3. CHECK: verify the result against my goal. If something doesn't fit, fix it and say what you changed. 4. REPORT: give the final answer, then list what you are least sure about. If you need information from me, ask before guessing. Don't claim to have done something you can't do (like buying or sending).
Exercise
Pick a small task you do every week (packing for a trip, a weekly grocery plan, answering the same type of email).
- Write its Goal in one sentence, including what “done” looks like.
- List the Plan steps you do yourself.
- Circle the steps an intern could Act on, and the ones you must keep (money, private data, final decisions).
- Write what you'd Check before trusting the result.
Then run the prompt above on that goal and compare its plan with yours. What did it miss? What did it do better?
Key takeaways
- A chatbot answers; an agent works toward a goal in a loop: Goal → Plan → Act → Check.
- Agents are robot interns: useful, fast, and in need of supervision.
- The check step is where trust is earned.
Related: Try the Dinner Planner agent