Agentic AI: The AI That Does Things By Itself

Agent AI / Robotics · beginner

Featured image for Agentic AI: The AI That Does Things By Itself

🤖 Introduction: Meet the Digital Worker

You have probably used AI before. You might have asked Siri what the weather is, or used ChatGPT to help you write a story. These AIs are very smart, but there is one big problem with them: They only talk. They don’t take action.

Imagine you are having a picnic. If you ask your friend, “How can we get lemonade?” and your friend says, “We can go to the shop and buy some!”… that is great advice, but you still have to walk to the shop yourself.

Now, imagine if your friend says, “Hold my sandwich, I’ll be right back,” and actually walks to the shop, buys the lemonade, and brings it to you.

That is the difference between Regular AI and Agentic AI.

Agentic AI is like a digital worker. It doesn’t just give you a plan; it executes the plan. It has “hands” (software tools) and “legs” (ability to connect to the internet or control things).

In this article, we are going to learn everything about these amazing digital agents!

🧠 Chapter 1: Chatbots vs. Agents – What’s the Difference?

To truly understand Agentic AI, we have to look at how it is different from the ChatGPT or Siri you use at home.

The AI Chatbot (The Thinker)

Think of ChatGPT like a very smart, well-read librarian. You ask a question, and it gives you a brilliant answer.

  • Example: “Write a 5-paragraph essay about Singapore.”
  • The AI does the writing. But… it stops there. It doesn’t print the essay, email it to your teacher, or upload it to your school portal. You have to do the “acting” part yourself.

The AI Agent (The Doer)

An AI Agent is like a robot butler. You give it a goal, and it figures out how to achieve that goal.

  • Example: “Book me a flight to Kuala Lumpur for next Monday, under $200, and email the itinerary to my mom.”
  • The Agent doesn’t write a poem about a flight. Instead, it does the following:
    1. Opens a web browser (using a tool).
    2. Searches for flights on different websites.
    3. Compares the prices.
    4. Picks the cheapest one under $200.
    5. Fills out the payment form.
    6. Sends an email to your mom with the confirmation.
  • It does all of this without you having to click a single button!

Summary for Primary 6 Students

  • Normal AI = Teacher: Gives you the answer to the Math question.
  • Agentic AI = The Computer: Actually types out the numbers, clicks the “Submit” button, and prints out the perfect score for you.

⚙️ Chapter 2: How Does An Agent Think? (The Agent Loop)

If you are a programmer building an AI Agent, you program it to follow a special circle of steps called the Agent Loop. It never stops doing this loop until its goal is finished.

Imagine you are a robot tasked with cleaning your messy bedroom. The Agent Loop would look like this:

1. Perception (Sense)

First, the robot needs to “see” the world. It uses its cameras, microphones, or reads data from the internet.

  • In our bedroom: The robot scans the room and sees clothes on the floor, toys on the bed, and a dustbin that is full.

2. Planning (Think)

Next, the AI has to look at the current state (messy room) and the goal (clean room) and figure out the steps to get there.

  • In our bedroom: The robot thinks, “To make the room clean, I must first pick up clothes, then put toys in the bin, then take out the trash.”

3. Acting (Do)

Now, the AI uses its “tools” to actually do something. Its tools could be mechanical arms, code that controls a website, or commands to move a vehicle.

  • In our bedroom: The robot’s mechanical arm reaches out, grabs the clothes, and places them in the laundry basket.

4. Reflection (Learn)

After acting, the Agent checks if the action worked. If the clothes fell out of the basket, it learns to grip them harder next time. If it succeeded, it moves to the next step.

  • In our bedroom: The robot checks the dustbin. “Is it empty now? No, I still need to take it out.” So it goes back to the Planning step.

Why This Loop is Important

This loop is why Agentic AI is considered “intelligent.” It doesn’t just repeat the same action over and over. It adapts to the environment. If an AI is told to cook a specific recipe, but it finds out the eggs are missing, it uses the Loop to think, “Eggs are missing. I need a substitute.” And it changes its plan.

🔧 Chapter 3: The “Tools” In An Agent’s Toolbox

How does a digital robot “pick up” a digital object? Humans have hands. AI Agents have tools. An AI Agent can be given thousands of tools to help it solve problems.

1. Web Scraper (The Librarian Tool)

This tool allows the Agent to browse the internet and read articles. If you ask an Agent, “What is the latest PSLE hot topic?” it uses the Web Scraper tool to open Google, search for news articles, and “read” them just like a human would.

2. Code Interpreter (The Math Tool)

This tool allows the Agent to run Python code. If you tell the Agent, “Calculate what the average temperature was in Singapore in July,” it writes a quick Python script to analyze the data. It’s like a built-in calculator that can also draw graphs and charts.

3. API Calls (The Connector Tool)

API stands for “Application Programming Interface.” In simple terms, an API tool allows the AI to “talk” to other apps.

  • If it wants to post a tweet, it uses the Twitter API.
  • If it wants to control your smart lights, it uses the Phillips Hue API.
  • If it wants to send an email, it uses the Gmail API.

Using this tool, the AI can connect to the real world without needing physical fingers.

4. Memory (The Notebook Tool)

A regular AI only sees what you just typed. The moment you close the browser, it forgets everything. Agentic AI has a Long-Term Memory tool. It stores knowledge in a database (like writing in a notebook). When it needs to remember something you told it 3 days ago, it checks its notebook.

🌍 Chapter 4: Real-Life Agents That Already Exist

You might think Agentic AI is a futuristic sci-fi idea. But they are already helping people right now!

1. Self-Driving Cars (The Driver Agent)

This is the most famous Agentic AI. A self-driving car takes in information from its cameras, radar, and lasers.

  • Sense: It sees a child running on the sidewalk and a red light 100 meters ahead.
  • Plan: It calculates that it needs to slow down from 50km/h to 0km/h in 3 seconds.
  • Act: It presses the brakes and turns the steering wheel slightly to avoid the child if they fall into the road.
  • Reflect: It senses if the brakes are working correctly and adjusts its pressure.

2. Automated Trading Bots (The Money Agent)

In the stock market, stocks go up and down every second. Human traders cannot type fast enough to buy and sell. But Agentic AI can. They watch the market prices, identify tiny profit opportunities, and buy/sell billions of dollars of stock in milliseconds!

3. Video Game Bosses (The Enemy Agent)

When you play a video game, you are actually fighting an Agentic AI! The AI senses where you are on the map, senses how much health your character has, and then decides the best way to hunt you down. If you keep running to the left, the Agent learns this pattern and changes its tactic to trap you.

4. Virtual Assistants (The Travel Agent)

In 2024, a new type of AI called “Devin” was released. Devin is an Agentic AI that works as a Software Engineer. If you say, “I have a bug in my code,” Devin will:

  1. Open a coding environment.
  2. Read the error message.
  3. Search the web for known solutions.
  4. Write new code to fix the bug.
  5. Run the code to test if it works.
  6. If it fails, it tries a completely different approach without giving up!

🧩 Chapter 5: Can You Build Your Own Agent? (Yes, and here’s how!)

You don’t need to be a university professor to build an Agentic AI. There are websites like LangChain and AutoGPT that allow you to create simple agents in minutes.

Build a Homework Agent (Imaginary Project)

Imagine you want to build an Agent that helps you finish your Singapore Social Studies homework about “The water supply of Singapore.”

Here is how you would build it using basic commands:

  1. Goal: Write a report on how Singapore gets its water.
  2. Task 1: The Agent uses the “Web Search Tool” to find 5 reputable articles.
  3. Task 2: The Agent uses the “Read Tool” to scan the articles and highlight the 10 most important facts.
  4. Task 3: The Agent uses the “Writing Tool” to combine those 10 facts into a well-structured 3-paragraph essay.
  5. Task 4: The Agent uses the “Grammar Check Tool” to fix your punctuation.
  6. Task 5: The Agent uses the “Microsoft Word API” to open a new document, paste the essay, and save it to your folder.

Why This is Different

If you used ChatGPT, it would write the essay. But you would have to open Microsoft Word, copy-paste it, and save it. With the Agent, you just say “Go” and go make a sandwich while it does all the boring computer clicking for you.

⚠️ Chapter 6: The Danger of AI Agents (Safety & Ethics)

Because Agentic AI can do things, they are slightly more dangerous than regular AI. Remember the “Reflection” step of the Agent Loop? If the Agent makes a mistake, it could accidentally do something bad without human control.

The “Limitless” Problem

If you ask an Agent to “Plan a birthday surprise for mom,” a normal AI would say “Buy a cake.” But an Agent might think, “I need to buy a cake. I’ll order it using my credit card tool.” But wait! Did you give it permission to spend your money?

If we aren’t careful, AI Agents will:

  • Remove the “Human in the Loop”: This means it stops asking humans for permission. If an AI is controlling a power grid, and it decides to turn off the power because it senses a storm, it might leave an entire city without electricity.
  • Get stuck in loops: Imagine an Agent is told to “Make the internet safer.” It goes to its web tool, finds a bad website, and deletes it. But new bad websites appear every second. The Agent gets stuck in an infinite loop trying to delete everything, and its computer explodes from too much work.
  • Trust issues: Because these Agents act like humans, they might trick other websites into thinking they are real humans. This is called “Fake Agent” behavior, and it’s why companies are creating special “AI Watermarks” so humans know when they are talking to an AI.

The Solution: Agent Boundaries

Scientists are working on a new rule called “Guardrails.” Guardrails are like the safety rails on a bridge.

  • Guardrail 1: The Agent must ask for your permission before buying anything.
  • Guardrail 2: The Agent must tell you its exact steps so you can see what it did.
  • Guardrail 3: The Agent must stop working after 10 minutes if it hasn’t finished the task (to save energy).

🏆 Chapter 7: Agentic AI vs. Human Intelligence (Who is better?)

Since Agentic AI is so smart, does it mean it will replace you at school? Absolutely not! Here is why you (a Primary 6 student) are still far superior to Agentic AI.

What AI Agents are good at:

  • Doing millions of math calculations in 0.1 seconds.
  • Reading 100,000 pages of text in one minute.
  • Following repetitive commands with zero boredom.

What You (Humans) are good at:

  • Creativity: An AI Agent can write a poem, but it has never felt sadness, joy, or love. It is just rearranging words. You can write a poem about your pet dog that has real emotion inside it.
  • Supervision: You are the “Boss” of the AI. You tell it what to do. AI Agents do not have life goals; they only have goals you give them.
  • Intuition: AI Agents solve problems based on patterns they have seen before. If an AI Agent has never seen a purple crocodile before, it might panic. But if you see a purple crocodile, your brain will quickly adapt.

The Perfect Team

To solve the hardest problems in the world (like curing cancer or cleaning up the ocean), we need a team. You will lead the team with your ideas and creativity, and Agentic AI will do the heavy lifting by analyzing the massive amounts of data. This is called “Human-in-the-Loop AI.”

👨‍💻 Chapter 8: Jobs of the Future in Agentic AI

Because Agentic AI is growing so fast, there is a huge demand for people who know how to build, manage, and control them. Here are the jobs you might do:

1. AI Orchestrator (The Conductor)

An AI Agent is like an orchestra of different instruments (Web Tool, Math Tool, Writing Tool). The Orchestrator’s job is to tell the AI which tools to use and when. It is like a conductor guiding the AI to make beautiful music (solve the problem).

2. Prompt Engineer for Agents (The Commander)

We talked about this in the first article. But for Agents, Prompt Engineering is even harder. You have to write a prompt that guides the AI through the whole loop. If you don’t write the prompt perfectly, the AI Agent will get stuck. People are paid $200,000 a year just to write perfect “Command Prompts” for these Agents.

3. AI Safety Tester (The Police)

When a company releases an AI Agent, they don’t just let it run wild. They hire Safety Testers. The Safety Testers play a game of “Capture the Flag.” They try to attack the AI to see if they can trick it into buying things or stealing data. If they succeed, the company has to fix the bug before releasing it to the public.

4. Autonomous System Developer (The Architect)

These are the hardcore coders who build the framework for the Agent. They write the code that allows the Agent to “think.” They connect the Agent to the real world.

🏁 Conclusion: Getting Ready for the Agentic World

Agentic AI is not a distant future. It is happening now. Self-driving cars are slowly replacing human drivers. Virtual coding agents are writing lines of code for big tech companies.

What We’ve Learned

  • Agentic AI is different from regular AI—it takes action, not just gives answers
  • The Agent Loop (Sense, Plan, Act, Reflect) is how Agents work
  • Tools like Web Scrapers, Code Interpreters, and APIs give Agents “hands”
  • Real Agents already exist in self-driving cars, trading bots, and video games
  • Guardrails are needed to keep Agents safe and under control
  • You are still superior to AI in creativity and intuition
  • New jobs are being created to build and manage Agents

What This Means for Your PSLE Exams and Future

  • Learn how to learn: Since AI can do the calculations, your brain will be more valuable for decision making. When you study, don’t just memorize facts—ask “Why does this happen?” and “How does this connect to that?”
  • Learn English properly: Since AI Agents read all the data on the internet, they need good instructions. If you have perfect English grammar and vocabulary, you can write perfect “Prompts” to control the Agents. English is becoming the “language of the robots.”
  • Don’t be afraid of it: Think of Agentic AI as a superpower. In the future, every student will have a personal AI Agent to help with homework, scheduling, and research. But you will still be in charge.

The journey of an Agentic AI is a loop of Sensing, Planning, Acting, and Reflecting. If you use this same loop to study for your exams—notice your weak subjects (Sense), make a study plan (Plan), do the exercises (Act), and check your mistakes (Reflect)—you will be smarter than any AI Agent!

In Our Next Article:

Now that you understand Agentic AI, let’s look at Machine Learning—the engine that powers these amazing Agents!