Today you hand a real, multi-step task to an AI agent and watch it work, start to finish.
Info
This is Part 3 of Your First AI Agent. New here? Start at Part 1. Up next: Writing Instructions an Agent Won't Misunderstand.
You're About to Cross a Line
Up to now, you have talked to AI. You typed, it answered. That's a chatbot.
Today is different. Today you hand off a whole task and let the AI take the steps to finish it. That's an AI agent at work.
It can feel like a big jump. It isn't, really. We're going to do one small task together, slowly, with you watching every move.
By the end, you will have delegated a real job from start to finish. No code. No setup headaches. Just you, the agent, and a task you can check.
Let's pick that task first.
Step 1: Choose a Safe, Low-Stakes Task
Your first run should be something you could check by eye in a few minutes. Nothing that sends money. Nothing that emails a real person. Nothing that deletes a file.
We want a task where a mistake costs you nothing but a little time.
Here are good first jobs:
- Research a topic and summarize the findings.
- Compare three products and recommend one.
- Turn a messy list of notes into a clean, sorted document.
- Draft a short plan from a goal you describe.
For this walkthrough, let's use one example you can follow along with: research three project management tools for a small team and recommend one.
It has multiple steps. The agent has to search, read, compare, and decide. That's exactly the kind of work an agent shines at.
Warning
Skip anything risky for now. No payments, no posting, no sending emails, no deleting. Keep your first task fully reversible. You can always check the next part on permissions once you're ready for bigger jobs.
Step 2: Write a Short, Clear Brief
An agent needs a brief. Think of it like the note you'd leave a new assistant.
A good brief answers three things. What's the goal? What does "done" look like? What are the limits?
You don't need fancy wording. You need to be clear. Here's a brief you can adapt:
Task: Research project management tools for a 5-person marketing team.
What I need:
- Find 3 popular tools that suit a small team.
- For each, list the price, the best feature, and one downside.
- Recommend one, and explain why in 2-3 sentences.
Limits:
- Only use information you can find from reputable sources.
- If you're unsure about a price, say so instead of guessing.
- Don't sign up for anything or enter my details anywhere.
Output: A short comparison and a clear recommendation.Notice the last limit. We told the agent what not to do. That single line keeps it inside the lines.
If writing briefs feels slow, our AI prompt generator can turn a plain sentence into a structured one like this. But a handwritten note works fine too.
Tip
The clearest part of any brief is the "done" line. Tell the agent exactly what a finished result looks like. "A 3-tool comparison and one recommendation" is far better than "help me choose a tool."
Step 3: Hand It Off
Now you paste the brief into your agent and let it go.
Where you do this depends on your tool. Many AI assistants now have an agent mode, a "tasks" feature, or a research button. The exact name varies. The idea is the same: you give the brief, then you step back.
When you hit go, resist the urge to add more. Let the agent read your brief and start. You'll watch what it does with what you gave it.
Open your agent tool and start a new task.
Paste your brief exactly as you wrote it.
If asked which permissions to allow, keep it to reading and searching only.
Press go, and then do not touch anything yet.
That last step matters. The first time, just watch. Don't jump in. We learn the most by seeing how the agent handles things on its own.
Step 4: Watch It Work
This is the part beginners skip, and it's the most useful part.
An agent works in a loop. It thinks about what to do. It takes one step. It looks at the result. Then it decides the next step. You'll see this play out on your screen.
For our example, you might see it:
- Search for popular project management tools.
- Open a few pages and read them.
- Note prices and features.
- Search again to fill a gap.
- Start writing the comparison.
Watching teaches you how the agent reads your brief. You'll see which words it took seriously. You'll notice where it guessed.
Keep an eye out for two things. First, looping: the agent repeats the same step over and over without progress. Second, drift: it wanders off into something you didn't ask for.
If you see either, you can stop it. Most tools have a stop or pause button. We'll go deep on spotting trouble in Part 6. For now, just know you're always in control.
Warning
If the agent starts doing something you didn't ask for, stop it right away. You are the supervisor. Stopping early is not failure. It's good judgment, and it costs you nothing.
Step 5: Review the Result Like an Editor
The agent finishes and hands you a result. You are not done. You're now the editor.
Don't read it as a fan. Read it as a skeptic with a checklist.
Ask yourself:
- Did it actually do all three parts of the task?
- Are the facts and prices believable, or did it guess?
- Does the recommendation match the comparison it wrote?
- Did it stay inside the limits I set?
This matters because agents can sound confident while being wrong. That confident-but-wrong habit is called a hallucination. The writing looks polished, so it's easy to trust. Slow down and check the claims.
For our example, you'd verify the prices yourself with a quick search. You'd make sure all three tools are real. You'd confirm the pick makes sense for a small team.
I read the agent's answer, it sounded great, so I used it as-is.
I checked the three prices, confirmed the tools were real, and caught one outdated number before sharing it.
If you want a second opinion on a written result, you can run it through our free prompt scorer or tighten a draft with the optimizer. But your own eyes are the main check.
Step 6: Decide What to Do With It
After your review, you have three choices.
Accept it. The result is good and the facts hold up. Use it.
Fix it yourself. Mostly good, one thing off? Edit that part by hand. Faster than starting over.
Send it back. A bigger miss means you adjust the brief and run again. Tell the agent what was wrong and what you want instead.
Sending it back is normal. It's not a failure. Often you'll spot one fuzzy sentence in your brief that caused the problem. You tighten that sentence, and the next run lands much closer.
Tip
When you send a task back, change one thing at a time. If you rewrite the whole brief, you won't know what fixed it. Small, single changes teach you what your agent needs to hear.
What You Just Learned
Step back and notice what happened. You picked a task. You wrote a brief. You handed it off, watched it run, and reviewed the result like an editor.
That's the full loop of delegating to an agent. Every future task follows this same shape, no matter how big.
The reviewing habit is the real skill. The agent does the legwork. You bring the judgment. That partnership is what makes agents safe and useful instead of scary.
Your first run won't be perfect, and that's fine. You're learning how this particular agent thinks. Each task makes your briefs sharper.
Info
Want a head start on briefs? Our template builder has structured starting points you can adapt for research, comparison, and planning tasks.
One thing kept coming up across these steps: the quality of your result depends on the quality of your brief. A fuzzy brief gives a fuzzy result. A clear one gives the agent a fair shot.
That's exactly where we go next. You'll learn how to write instructions an agent reads the same way you meant them, so you stop getting surprised by the output.
For now, go run one real task. Pick something small. Watch it work. Review it like an editor. You've got this.
Keep going
Next → Part 4: Writing Instructions an Agent Won't Misunderstand
Or see the full Your First AI Agent series.