Ask It Simply
Build With AI · Advanced · Lesson A1

What an AI agent is, and isn't

about 8 minutes. Level: Beginner. Nothing to install.

By Dontè

Watch the video

Lesson video, about 4 min 47 s. Captions are off by default. Press CC to turn them on. It starts muted. Press Sound to hear the lesson from the start. Press and hold the playing video (or hold the period key) to watch at 2x speed.
Show transcript

Hi, and welcome to Build With AI. This is Lesson A1. In the next few minutes you'll learn what an AI agent really is, why the smartest builders start simple, and you'll design your very first agent. No coding. Nothing to install. Let's go.

There are three kinds of AI helpers you'll hear about. A chatbot, a workflow and an agent. They sound alike, but they work very differently.

First, a chatbot. You ask, it answers. That's it. When you type a question into Grok and get a reply, that's a chatbot conversation.

Second, a workflow. Anthropic, the company that makes Claude, describes it like this. Quote: "Workflows are systems where LLMs and tools are orchestrated through predefined code paths." End quote. In plain words, someone wrote the steps down ahead of time, like a recipe. The AI follows the recipe.

Third, an agent. OpenAI's guide says, quote: "Agents can plan and complete tasks using tools, work with other agents, and maintain context across steps." End quote. So an agent gets a goal and some tools, and it decides its own next step.

Here's an easy way to remember it. An agent is AI, plus tools, plus a goal, plus a loop. A tool is something the AI can use, like a calendar or a calculator. The loop means it thinks, acts, checks the result, and repeats.

Picture a helper who is asked to find a good time for a family dinner. It thinks, "I should look at the calendar." It acts by checking the calendar. It checks, "Is Saturday free?" Then it repeats until it has an answer. That loop is what makes an agent an agent.

Now, here's one of the most useful ideas in all of AI building. Anthropic writes, quote: "we recommend finding the simplest solution possible, and only increasing complexity when needed. This might mean not building agentic systems at all." End quote.

Think about that. One of the top AI companies is saying, sometimes you don't need an agent at all.

Our tip: before you build an agent, ask yourself, "Could a simple checklist or one good prompt do this?" If the answer is yes, start there. You can always add more later.

Okay, let's design an agent. Open Grok. You can use grok.com or the Grok app. On the lesson page there's a block you can copy. Copy the whole thing.

Paste it into Grok. Let's read it together. The goal is to remind me about bills before they're due. It can use two tools, picked from a calendar, an email reader, a calculator or notes. And here's the most important line, the stop rule. It must stop and ask me before it does anything that spends money or sends a message.

We ask Grok for four things. A one-sentence job description. The two tools and why. The loop in four short steps. And one simpler option that might work without an agent. And we ask for plain words.

Here's Grok's answer. Let's check it. Did it keep the stop rule? Look for it in the plan. Did it give us a simpler option? Look for something like a calendar reminder for each bill. When we tried it, Grok suggested a notes list of your bills, with phone alarms set a few days early. If it's there, that's a great sign. It matches the "start simple" advice.

Your answer might look a little different from mine. That's normal. AI answers change from one try to the next. What matters is that the stop rule is there and the plan makes sense to you.

Want a second opinion? You can paste the same block into ChatGPT or Claude and compare. But Grok is our lead helper for this whole section.

Before we finish, the safety rule. Our tip: agents make mistakes. A human, that's you, approves anything that spends money or sends messages. Build that rule into every agent you design, starting today.

Quick check. Which one follows steps planned in advance? A workflow. What are the four parts of an agent? AI, tools, a goal and a loop. And what should you try before building an agent? The simplest thing that works.

Nice work. You just designed your first agent. In Lesson A2, you'll make your first real call to Grok from a browser notebook, with nothing to install. See you there.

What you'll learn

Step 1: Three kinds of AI helpers

A chatbot answers you. You ask a question, it replies, and that's it. Grok in a chat window works this way.

A workflow follows steps someone planned ahead of time. Anthropic, the company that makes Claude, puts it this way: "Workflows are systems where LLMs and tools are orchestrated through predefined code paths." (Anthropic, "Building effective agents") In plain words, the steps are written down in advance, like a recipe.

An agent is given a goal and some tools, and it decides the next step itself. OpenAI's guide says: "Agents can plan and complete tasks using tools, work with other agents, and maintain context across steps." (OpenAI, Agents guide)

A simple way to remember it:

Agent = AI + tools + a goal + a loop (think, act, check, repeat)

New words: - Tool: something the AI can use, like a calculator, a calendar or a web search. - Loop: the AI tries a step, checks the result, then picks the next step.

Step 2: Start simple

Here is one of the most useful lines in AI building. Anthropic writes: "we recommend finding the simplest solution possible, and only increasing complexity when needed. This might mean not building agentic systems at all." (Anthropic, published Dec 19, 2024)

Our tip: before you build an agent, ask "Could a simple checklist or one good prompt do this?" If yes, start there.

Step 3: Try it (copy and paste, no setup)

Open Grok (grok.com or the Grok app). Copy this whole block, paste it in, and fill in the blanks.

Help me design a simple AI agent on paper. Do not write code yet.

My goal: remind me about bills before they are due.
Tools it may use (pick 2): calendar, email reader, calculator, notes.
Stop rule: it must stop and ask me before it does anything that spends money or sends a message.

Please give me:
1. A one-sentence job description for the agent.
2. The 2 tools it needs and why.
3. Its loop in 4 short steps (think, act, check, repeat).
4. One simpler option that might work without an agent.
Use plain words a 12-year-old would understand.

What to look for in Grok's answer: - Did it keep your stop rule? - Did it suggest a simpler option? That's a good sign, and it matches the "start simple" advice above.

Want a second opinion? You can paste the same block into ChatGPT or Claude and compare.

Grok's real answer (our test run)

We ran this exact prompt through Grok (model grok-4.7) on Oct 30, 2026. Grok's answers change from run to run, so yours will differ. Quoted as Grok returned it, except where we mark [trimmed] and where we swapped dashes for commas to match our style.

  1. This agent looks for your bills and reminds you a few days before each one is due.

  2. It needs two tools: - Email reader, so it can find bill emails and read the due date and amount. - Calendar, so it can put a reminder on a day before the bill is due.

It does not pay bills or send messages. If it ever wants to do either of those, it must stop and ask you first.

  1. Its loop: - Think: Look at new bill emails and note what is due and when. - Act: Add a calendar reminder a few days before the due date. - Check: Make sure the reminder is early enough and that nothing was paid or sent without your OK. - Repeat: Check again when new mail arrives.

  2. A simpler option: keep one notes list of your bills and their due dates, and set phone alarms yourself a few days early. No agent needed.

Notice it kept the stop rule and gave a simpler option.

Check yourself

  1. Which one follows steps planned in advance: a chatbot, a workflow or an agent? (Answer: a workflow)
  2. What are the four parts of an agent? (Answer: AI, tools, a goal, a loop)
  3. What should you try before building an agent? (Answer: the simplest solution that works)

Caution box Our tip: Agents make mistakes. A human approves anything that spends money or sends messages. Build that rule into every agent you design.

Builds on

Sources

Creator credits

Keep going

NextA2 →TrackBack to Build With AI →