Build your own AI agent
An AI agent is a computer helper that can think in steps and do a job for you. Here, you will build one yourself, no code needed. Read what's ahead below, then start whenever you're ready.
Listen to this briefing
Build Safe AI Agents Without Code
What this tool does
An AI agent is a computer helper that can think about a job in steps, then try to do that job on its own. Real companies are starting to use AI agents to help with real work. But before anyone switches one on, they build it carefully and check that it is safe first.
On this page, you will build a practice AI agent yourself, then run a safe, practice test on it. You will not type any code. You will just answer simple questions. It is free and takes about 5 to 10 minutes. Your answers are kept in this browser so you can come back to them, and nothing is ever sent to anyone.
What you will do: 7 short steps
- 1Name it. Give your agent a name, like Robo or Sunny.
- 2Give it a job. Pick one clear job for it to do.
- 3Give it tools. Choose what it is allowed to use, like a web search.
- 4Set its rules. Choose rules it must always follow.
- 5Test it. Write it a task and watch it think, step by step.
- 6Check it is safe. Answer 4 safety questions real companies ask before they launch an agent.
- 7Get your record. Open a full report of everything you built, ready to print or save as a PDF.
How to use this tool
- Go one step at a time. Fill in each step, then press Next.
- Press Back, or a finished step in the progress bar, any time to change something you already picked.
- The Next button turns on once you have filled in enough for that step.
- Every step has a short "Why this step matters" box. Read it if you want to know more.
- This is a practice space with no real AI behind it. Your answers stay in this browser until you press Start over. Nothing is sent to anyone.
Step 1 of 7: Give your agent a name
Giving your agent a name makes it feel real, like someone you can talk to. A name does not change what your agent can do. It just makes it easier for you to think about your agent while you build it.
Pick a job for it
Pick a job below. It will fill in a starting goal for you. You can change the words after.
A job is your agent's one big goal. Real AI agents work best with one clear job, not five. A clear goal helps your agent know what "done" looks like, and stops it from wandering off into things you did not ask for.
Give it tools
A tool is something your agent is allowed to use to do its job. Pick at least one.
A real AI agent can only do things it has tools for. If a task needs a tool your agent does not have, it cannot finish the job. Pick tools that match the job you gave it in Step 2.
Set its rules
A rule is something your agent must always do, or never do. Pick at least one.
Rules keep your agent safe. Even a smart agent can make mistakes, or try to do something you did not want. Rules are like a fence: they keep your agent on safe ground, even when a task is tricky or unclear.
Write it a task
Your Agent
Pick a job and give it a name to get started.
Tools: none chosen yet
Rules: none chosen yet
This is called a "prompt." It is the exact job you are giving your agent to try, right now.
- ✓Say exactly what you want. "Write a thank-you email" works better than "help with email."
- ✓Add the important details. Names, numbers, and dates help your agent get it right.
- ✓Keep it short and clear. Use simple sentences. One idea at a time.
- ✓Ask for one task, not five. A single clear ask works better than a long list.
- ✓Say how it should sound, if that matters. Example: "in a friendly voice" or "in three short bullet points."
"help with the customer thing"
"Write a short, kind reply to a customer whose order is 3 days late. Tell them it is on the way, and offer a 10% discount code."
Write a task above, then press Test My Agent to see how it thinks.
This is a practice tool that teaches the ideas behind AI agents using simple, built-in rules. It does not call a real AI model, so every "answer" it gives is a scripted demonstration. Everything you type stays in your browser. It is never sent to Terence Kok or anyone else.
Make it safe to launch
Before any real AI agent starts working, grown-ups check a few important things first. This is called governance. Check the boxes that are true for your agent.
Real companies follow rules before letting an AI agent loose in the world. Singapore's government agency, IMDA, wrote a framework just for agents like this one. It says: know the risks first, keep a person in charge, build in safety checks, and watch the agent after it starts. These four checks are a simple version of that.
For grown-ups: this maps to Singapore's IMDA Model Governance Framework for Agentic AI (2026): the world's first governance framework written specifically for agentic AI.
Your AI system record
Real companies write down how they built and checked their AI agent. This is called an AI management system record. Here is yours, in plain English.
Not written yet.
Not written yet.
Not written yet.
If a test shows a missing tool or a broken rule, add it, then test again. Real teams do this every time they change their agent.
Not written yet.
ISO/IEC 42001 is the world's standard for managing AI responsibly. It asks every organisation to write down the same five things you just saw: the goal, the controls, how it is checked, how it improves, and a record of the system itself. Big companies get audited on this. You just did a simple version of it.
This safety check and record are simplified for learning. They are based on real frameworks: Singapore's IMDA Model Governance Framework for Agentic AI and ISO/IEC 42001, but finishing them here does not mean a real agent is certified or compliant. A third framework worth knowing, the US NIST AI Risk Management Framework, covers similar ground through four functions (Govern, Map, Measure, Manage) rather than clauses; see it in the Standards Reference. For a real organisation, use the full ISO 42001 AIMS Readiness Checklist and TRACE Agent Evaluation tools, or read about the IMDA framework in full.
Four words to know
Every AI agent, even a real one, is built from these four parts. Learn them here, and you will understand how any AI agent works.
Goal
What you want your agent to do. Every agent needs one clear job. Without a goal, an agent does not know what it's for.
Tools
What your agent is allowed to use to do its job, like a web search or a calculator. An agent can only use tools you give it.
Rules
What your agent must always do, or must never do. Rules keep an agent safe and helpful, even when a task is tricky.
Prompt
The exact task you type in, in your own words. A clear prompt helps your agent understand exactly what to do.
The order your agent uses its four parts
Schematic of the practice test on this page. There is no real AI behind it: the test follows these checks in this order every time.
I built this as the on-ramp, the version for someone who has never touched an AI agent and just wants to understand what one is before the jargon shows up. What you type stays in your browser, and there's no real model running behind it. What I hope sticks is the shape: a goal, some tools, a few rules, and a person checking the work before it ships. That shape changes little as systems get more advanced; only the consequences grow.
Last updated 29 Sept 2026
More in Agent labs
- TRACE Agent EvaluationIs this task safe to hand to an autonomous agent at all?
- Expert AI Agent BuilderWould our guardrails hold against a real attack?
