gluntyacademy

glunty academy

Agent Lab - Lesson 1

How agents actually work

Most people think AI is magic. It is not. An agent is a LOOP - think, act, look, repeat - and you are the one who runs it. You found the lab. Now learn the trick.

0 of 8 done
  1. Finish the step above first
    01

    The secret nobody tells you

    An AI agent is not a brain. It is a <b>loop</b>.

    Goal, then think, then act, then look at what happened, then go again. That is the whole trick behind every "AI agent" you have heard about.

    You are not the passenger. You are the <b>operator</b> who starts the loop and steers it.

  2. Finish the step above first
    02

    Learn the four moves

    Every agent repeats the same four moves:

    <b>1. Goal</b> - what are we trying to do. <b>2. Think</b> - what is the next step. <b>3. Act</b> - use a tool or write something. <b>4. Check</b> - did that work? If not, loop back to Think.

  3. Finish the step above first
    03

    Give it a real goal

    A goal has a finish line. "Help me" is a wish. "Make a 5-question quiz about volcanoes and grade my answers" is a goal.

    Write a goal with a clear finish line.

  4. Finish the step above first
    04

    Run one turn of the loop

    Open a chat AI. Paste the prompt. It makes the agent PLAN first, then do only step one and stop - so YOU stay the operator.

    what will really happen

    The AI will want to sprint ahead and do everything at once. Making it go one step at a time is how you stay in control - that is the operator move.

    Open ChatGPT (new tab)
    You are my agent. Goal: {goal}. First, list the steps you will take. Then do ONLY step one and stop, so I can check it before you continue.
  5. Finish the step above first
    05

    Give it tools

    An agent gets powerful when it can USE things: search the web, run code, read a file, open a link.

    What tools would make YOUR agent stronger at its goal?

  6. Finish the step above first
    06

    Make it check itself

    This is the part beginners skip. A real agent checks its own work before it trusts it.

    How could your agent tell if it did the job right?

    hold on - judge it, do not celebrate
    • Would you have caught a wrong answer, or just believed it?
    • What is the test that proves it worked?
  7. Finish the step above first
    07

    Plan for the miss

    When the check fails, a good agent does not give up - it loops back and tries a different step.

    What should yours do when the check says "nope"?

  8. Finish the step above first
    08

    Read back your agent

    Here is your agent, as a plan an operator would actually run. Copy it and keep it.

    Do the steps above and your your agent plan shows up here.

The class wall - share yours

Loading the class wall...

AGENT

You think like an operator now

You just ran the loop that every AI agent runs. Keep this plan. Next in the lab: turn your loop into a reusable helper you get to keep.

Back to the academy Your progress
Embedded tool from glunty.com