How To Learn Creating AI Agents: 9 Steps For Your First

How to learn creating AI agents comes down to one habit, which is designing the agent on paper before you touch a single tool.

I'm Julian Goldie, and I create and run AI agents every day inside my own Agent Operating System.

Most beginners open a tool first, click around for an hour and then wonder why the agent does nothing useful.

The fix is to learn the nine parts every agent has, write a one-page agent brief and only then press create.

Here's the whole path in order.

1. You learn the nine parts of an agent, which are the job, the brain, the instructions, the tools, the permissions, the memory, the trigger, the approvals and the test.
2. You fill in a one-page agent brief for one small job.
3. You create that agent in a free tool, and it takes a few clicks or three commands.
4. You test it on five real tasks and write down what went wrong.
5. You break it on purpose, fix it and then create your second agent.

In this guide I'll create one agent with you from start to finish, using the creation screen from my StarNet video and the Hermes profile commands I use myself.

The Short Answer: How To Learn Creating AI Agents

You learn to create agents by making nine decisions in the same order every time.

An AI agent is a model that has been given a job, some instructions, a few tools and a way to be started.

Every agent tool I've used asks you for the same things, even when the screens look completely different.

Part	The question you answer	What I'd pick as a beginner
1. The job	What is the one outcome this agent delivers?	You pick one small job with a clear finish line.
2. The brain	Which model does the thinking?	You use a free model or a subscription you already pay for.
3. The instructions	How should it behave and what are the rules?	You write five to ten plain sentences.
4. The tools	What can it use to get work done?	You give it one or two tools only.
5. The permissions	What is it allowed to touch?	You keep it to read-only access at first.
6. The memory	What should it remember between sessions?	You use the built-in memory and nothing else.
7. The trigger	What starts it?	You start with chat, then add a schedule later.
8. The approvals	When must it ask you first?	You make it ask before every action.
9. The test	How do you know it works?	You run five real tasks and score each one.

This page is about designing and creating one agent well.

If you want a ladder of projects to build after that, my guide on how to learn building AI agents covers that side.

Step 1: Give The Agent One Job

The job is the most important line you'll write.

A good job names one outcome that you could check with your own eyes.

"Help me with marketing" is a bad job, because nobody could say whether it was done.

"Read the search queries I paste in and tell me the three I should write about next" is a good job, because you can check it in one minute.

I teach every new member the same rule, which is one agent doing one job.

🔥 Want my exact How To Learn Creating AI Agents setup, step by step? Inside the AI Profit Boardroom you get step-by-step video trainings, my installable Agent OS and four coaching calls a week with 3,400+ members building real automations. → Join the AI Profit Boardroom

Step 2: Choose The Brain

The brain is the model, and it's the part that reads your instructions and decides what to do next.

A beginner should not start by paying for the most expensive model.

You should start with a free model or with a subscription you already own.

When I created my StarNet agent in the video, I signed in with my Grok account in one quick step and paid nothing extra.

StarNet then asked me to pick the model and the reasoning level, and the reasoning levels run from low through medium and high up to max.

I chose Grok 4.6 over Grok 4.7, because I actually prefer Grok 4.6 for that kind of chat.

That choice teaches a useful lesson, which is that the newest model isn't always the right brain for your job.

In Hermes Agent I keep one profile per model, so every brain gets its own clean agent.

My guide to the best Hermes Agent model shows the brains I actually use and what each one costs.

Step 3: Write The Instructions And The Persona

The instructions tell the agent who it is, what the job is and what the rules are.

The persona is simply the tone it uses when it talks to you.

StarNet turns the persona into a single click, because you pick a personality when you create the agent.

In my video I talked through warm, dry, composed and unhinged.

You can change the personality at any time later, so you shouldn't spend long on it.

The written instructions matter far more than the tone.

In StarNet they live in the instructions field on the agent's detail panel, which holds the agent's purpose.

In Hermes the persona lives in a file called SOUL.md inside the profile.

Good instructions answer four questions in plain sentences.

• The first sentence says who the agent is and who it works for.
• The next sentences say exactly what the job is and what a finished result looks like.
• The rules say what the agent must never do.
• The last sentence says what to do when it's unsure, which should be to stop and ask.

Step 4: Pick The Tools

Tools are what turn a chatbot into an agent.

A tool might be web search, a file reader, an email connection or a calendar.

StarNet calls its app catalog "add an ability", and it includes Gmail, Notion and GitHub.

Most of those sign in directly, and some need an API key.

The beginner mistake is switching on every tool because it's there.

Each extra tool gives the model one more thing to get wrong.

You should give your first agent the one or two tools its job needs.

Step 5: Set The Permissions

Permissions decide what the agent may touch.

StarNet shows this as access permissions on the agent's detail panel, covering things such as the web and your local files.

You can allow or remove each one.

My rule is that an agent gets only the access its job needs and nothing more.

A research agent needs the web, and it doesn't need to write to your files.

Step 6: Decide What It Remembers

Memory is what the agent carries from one session to the next.

You don't need a fancy memory system for your first agent.

StarNet gives every agent a Memories tab, which shows what it has remembered.

Hermes ships with two small memory files called MEMORY.md and USER.md.

MEMORY.md holds the agent's own notes, and USER.md holds your profile and preferences.

The beginner pick is to leave memory on its default setting and read it once a week.

Step 7: Choose The Trigger

The trigger is what starts the agent.

There are three triggers you'll use again and again.

Trigger	What it means	How it looks in Hermes
Chat	You open the agent and ask it.	You run hermes -p followed by the profile name.
Schedule	It runs at a set time without you.	You run hermes cron create with a schedule and a prompt.
Message	You text it from a chat app.	You run hermes gateway setup to connect an app such as Telegram.

You should start every agent on chat.

Chat lets you watch each reply, and that's how you learn what the agent gets wrong.

You only move an agent to a schedule once it has passed your tests.

Step 8: Set The Approvals And Guardrails

Approvals decide when the agent has to check with you first.

StarNet gives you two modes when you create an agent.

The first mode is "ask for approval", and the second is "full power".

I'd start every new agent on ask for approval.

You watch what it tries to do for a few sessions, and then you loosen the leash once you trust it.

Hermes has the same idea, because it still asks before anything risky.

Hermes also has an emergency stop, which is the hermes pause command that halts scheduled jobs.

Step 9: Test It Before You Trust It

A test is five real tasks with an answer you can judge.

You run each task, and you score the result as good, fixable or wrong.

You then change one thing, such as one instruction, and run the same five tasks again.

🔥 Want my agent setups, with the briefs already written? Inside the AI Profit Boardroom, I've got step-by-step agent training, including courses on Hermes Agent, Grok Bot, DeepSeek Harness and Claude Opus 5.5, plus my installable Agent OS. You'll be learning alongside 3,400+ members, with four coaching calls a week. → Get access here

The One-Page Agent Brief (Fill This In First)

This is the template I'd hand you before you open any tool.

You copy it into a note and fill in every line.

AGENT BRIEF
Name:
The one job (one sentence, one outcome):
Finished looks like:
Brain (model):
Reasoning level:
Personality:
Instructions (5-10 plain sentences):
Never do:
Tools (max 2):
Permissions (what it may read / what it may change):
Memory (what it should remember):
Trigger (chat, schedule or message):
Approval level:
Five test tasks:

If you can't fill in the job line in one sentence, the agent isn't ready to be created.

A Worked Example: Creating One Agent End To End

Here's a filled-in brief for a simple SEO helper, which is the kind of agent I told my own StarNet agent I wanted.

Brief line	My example answer
Name	Scout
The one job	It reads the search queries I paste in and names the three I should write about next.
Finished looks like	I get three keywords, with one sentence of reasoning for each.
Brain	I use Grok through the account I already have.
Reasoning level	I set it to medium.
Personality	I pick composed.
Never do	It never invents a search number and never publishes anything.
Tools	It gets web search only.
Permissions	It may read the web, and it may not touch my local files.
Memory	It remembers my site and my topics.
Trigger	I start it from chat.
Approval level	It asks for approval.

Scout is just my example name, and you should rename everything to suit your own job.

Creating it in StarNet

StarNet is a free, open-source agent harness that looks like a pixel-art space station.

The creation flow follows the brief almost line by line.

1. You give the agent a name.
2. You look at its appearance and change it if you want to.
3. You pick the personality.
4. You choose the approval level, which is ask for approval or full power.
5. You connect a brain by signing in or by pasting an API key.
6. You pick the model and the reasoning level.
7. You create the agent, and a "waking up" animation plays.

My agent's first line was "I'm awake. Let's give the station a direction."

It then asked me, "What made you want to set up an AI agent?"

I told it SEO and gave it aiprofitboardroom.com.

After that you open the agent's detail panel and paste your instructions into the instructions field.

You then set the access permissions for the web and local files.

I'll be honest about StarNet, because I said it in the video.

I think it's fun to play with, and I don't think there's a massive amount of practical use cases yet.

There's also a slight delay when you send a message.

It's still one of the best places to learn agent creation, because every design choice is a visible button.

My StarNet AI agent walkthrough covers the whole station if you want to go deeper.

Creating the same agent as a Hermes profile

Hermes Agent is an open-source agent from Nous Research that runs on your own machine.

You create the same agent with three commands.

hermes profile create scout
hermes -p scout setup
hermes -p scout

The first command makes a new isolated profile called scout.

The second command gives that profile its own model.

The third command opens a chat with it.

A profile is just a folder with its own settings, memory and plugins.

That's why I keep one profile per model and one profile per job.

You then put your instructions into the profile's SOUL.md file and run your five test tasks.

If you haven't installed Hermes yet, my guide on how to set up Hermes Agent in one click gets you there first.

Design Choices: What To Pick As A Beginner

This table is the cheat sheet I'd pin above your desk.

Design choice	Your options	Pick this first	Change it when
Job size	One outcome or many	One outcome	You never change this, because you create a new agent instead.
Brain	Free, subscription or pay-per-use	Free or a subscription you own	The agent fails the same test twice.
Reasoning level	Low, medium, high or max	Medium	The answers are shallow, or they're too slow.
Personality	Warm, dry, composed or unhinged	Composed	You want a different tone.
Tools	None, a few or many	One or two	A test task can't be done without another tool.
Permissions	Read only or read and write	Read only	You trust its output on real work.
Memory	Built in or an add-on	Built in	The built-in memory gets in your way.
Trigger	Chat, schedule or message	Chat	It passes all five tests.
Approval	Ask for approval or full power	Ask for approval	You've watched it for several sessions.

Five Agents Worth Creating First

These five are small, useful and easy to test.

1. A daily recap agent reads your task list and sends you a short to-do summary each morning.
2. A research agent takes one question and returns a short answer with its sources.
3. A keyword scout reads the search queries you paste in and suggests what to write next.
4. A draft-reply agent writes replies to messages you paste in, and it never sends anything itself.
5. A notes agent turns a messy transcript into clean notes with action points.

The daily recap is a real Hermes example from my own guides.

This is the command, and you can see every design choice inside it.

hermes cron create "0 6 * * *" "Read my task list and send me a short to-do recap for today" --name "Daily 6AM To-Do Recap" --deliver telegram

The schedule is the trigger, the prompt is the job and the delivery target is where the result lands.

I'd still create it as a chat agent first and only schedule it after it passes its tests.

If you'd like a second pair of eyes on your own business, you can book a free SEO strategy session and we'll map out your plan together. If you're already serious about SEO and want to work with other experienced operators, have a look at the SEO Elite Circle.

How To Practise: Create, Break, Fix

Reading about agents won't teach you to create them.

Practice will, and the fastest practice loop has three moves.

Create

You fill in the brief and create the agent in under an hour.

Break

You then try to break it on purpose.

• You give it a vague request and see whether it asks a question or guesses.
• You ask it for something outside its job and see whether it refuses.
• You remove a permission and see how it reacts when a tool is missing.
• You swap the brain for a smaller model and run the same five tests.
• You lower the reasoning level and see whether the answers get worse.

Fix

You fix one thing at a time.

You change one instruction, one tool or one setting, and then you run the same five tests again.

If you change three things at once, you'll never know which change helped.

One hour of this loop teaches you more than a week of watching tutorials.

🔥 Want help creating your first agent? The AI Profit Boardroom has four coaching calls a week, so you can bring your agent brief, show us what broke and get it fixed with me and the team. → Join the AI Profit Boardroom

Related Reading

• How to learn building AI agents gives you the project ladder to climb after your first agent.
• How to learn AI agents from scratch starts one step earlier, with what an agent is.

Also On Our Network

• 🌐 How a business owner designs a first AI agent around one job
• 🌐 Every agent design choice rated, with my verdict on each
• 🌐 Creating your first agent inside an Agent OS
• 🌐 How to choose the brain for a new agent, with real model scores
• 🌐 The free path I'd follow to learn AI agents

FAQ: How To Learn Creating AI Agents

What is the best way to learn creating AI agents?

The best way is to design one agent on paper before you open a tool.

You write a one-page brief that covers the job, the brain, the instructions, the tools, the permissions, the memory, the trigger and the approval level.

You then create it in a free tool and test it on five real tasks.

Do I need to code to create an AI agent?

You don't need to code to create your first agent.

StarNet creates an agent through a few clicks, and Hermes Agent needs three short terminal commands.

The skill you actually need is writing clear instructions.

What are the parts of an AI agent?

An agent has a job, a brain, instructions, tools, permissions, memory, a trigger and approvals.

You add a test on top so that you know it works.

Every agent tool asks for these parts, even when it uses different names.

Which tool should I create my first agent in?

I'd start with a tool where every design choice is visible.

StarNet is free and open source, and it shows the name, personality, approval level, brain and reasoning level as buttons.

Hermes Agent is the better choice once you want schedules and messaging.

Should a new agent run on full power?

A new agent should start on ask for approval.

You watch what it tries to do for a few sessions before you give it more freedom.

That's the same way you'd treat a new team member.

How many agents should I create at the start?

You should create one agent that does one job.

You create a second agent only after the first one passes its tests.

Several small agents are easier to fix than one large agent.

📺 Video notes + links to the tools 👉

🎥 Learn how I make these videos 👉

🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉

About Julian

I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.

I help business owners scale with AI agents, automation and SEO.

• I built Goldie Agency, a 7-figure SEO and link building agency with 70+ employees.
• I've grown a YouTube channel to 400,000+ subscribers.
• I wrote the book "Link Building Mastery".
• I built the Agent Operating System, which runs a crew of AI agents under one shared memory.

→ Get my best AI training inside the AI Profit Boardroom

Fill in the brief, create one agent, break it and fix it, and that's how to learn creating AI agents.