AI WORLD HQ
Picture four tiny helpers sitting at a desk: one finds articles, one sums them up, one writes a draft, and one sends the email. That's an AI agent team. None of them are programmers, and neither are you — you just give each helper clear instructions and a place to leave notes. This article walks through what that looks like in practice, using one everyday example (a weekly newsletter) you can actually try.
Before you start
A few things worth knowing before you set anything up.
- No coding required. Every step below uses tools with normal point-and-click menus. If you can write a Google Doc, you can do this.
- You'll need a free or paid AI account. Most beginners start with ChatGPT, Claude, or Gemini. A free tier works; paid plans ($20/month or so) just give you more messages and longer chats.
- You'll need a place for your agents to "leave notes". A free Google Doc, a Notion page, or even a shared folder on your computer is enough.
- Pick one task to start with. Trying to automate your whole life on day one is the fastest way to give up. Pick one annoying task — like a weekly newsletter, a long email, or a list of links — and start there.
What even is an AI agent?
An AI agent is a chatbot (a chat-style AI you type questions to) that's been given a specific job and a specific set of instructions. Instead of answering any question, it only does one thing — like "summarize the article I'm pasting in" or "draft a reply to this email in a friendly tone".
A team is just several agents, each with a different job, passing their work to each other through a shared document. The research agent doesn't try to write. The writing agent doesn't try to research. They stay in their lane.
This pattern — one job per agent, plus a shared space for handoffs — is the whole idea. You don't need any special software to make it happen.
Step 1 — Pick one task and break it into 3 or 4 jobs
The trick to a working team is making each job small and obvious. For our newsletter example, the jobs are: research → summarize → write → send. Four steps, each one a sentence.
Pick a task from your own life. Maybe it's "every Sunday, plan the family's meals for the week". The jobs could be: suggest meals → build shopping list → check what's already in the fridge → write the meal plan.
💬 Copy this prompt into your AI chat: "Help me break the task of [your task here] into 3 or 4 small, single-purpose jobs. Each job should be something one AI agent could do on its own."
You'll know it worked when you can read each job out loud in one breath, and no job overlaps with another.
Step 2 — Create one AI chat per job
For each job, open a new chat in your AI tool of choice (ChatGPT, Claude, or Gemini all work). Name the chat after the job — for example, "Newsletter — Research" — so you don't mix them up.
In each chat's first message, give the agent its job and its rules. This is the instruction that tells the agent what it is and what it isn't allowed to do. Without it, the agent will try to do everything at once.
💬 Copy this prompt into your first message in each new chat: "You are the Research Agent for my weekly newsletter. Your only job is to find articles. You do not write, summarize, or edit anything. When I ask, return the top 5 items as a bulleted list with links and one-sentence summaries."
You'll know it worked when you ask the agent a question and it stays strictly within its one job — even if you try to tempt it to do more.
Step 3 — Give every agent a place to leave notes
This is the part most guides skip, and it's the part that makes the team actually work. Each agent needs somewhere to put its output so the next agent can read it.
The simplest option: open a free Google Doc and make one section per job — "Research", "Summary", "Draft". Each agent's output gets pasted into its section. The next agent's input is whatever's in the previous section.
If you already use Notion, one page with four headings works the same way. A simple text file on your computer works too.
💬 Copy this structure into a new Google Doc:
Research: (paste here) Summary: (paste here) Draft: (paste here) Final: (paste here)
You'll know it worked when you can point at one heading and say which agent fills it in and which agent reads from it.
Step 4 — Run the team once, by hand, end to end
Before you try to automate anything, run the team manually so you can see where it breaks. Open the Research agent's chat, give it today's task, and paste its output into the Research section of your shared doc. Then open the Summarize agent, tell it to read the Research section and produce a Summary, and paste that in. Repeat for Write, then Send.
This sounds slow. It is — the first run takes 20 to 30 minutes. The point is to find the spots where one agent's output isn't quite what the next agent needs, so you can fix the prompt.
💬 Copy this prompt into your Summarize agent: "Read the Research section of the linked Google Doc. Return 5 bullet points, each one sentence long, capturing the most useful fact or update."
You'll know it worked when the final output — the newsletter draft, the meal plan, whatever you set out to build — actually looks like something you'd send or use.
Step 5 — Fix the gaps by editing prompts, not code
Almost certainly, one agent's output won't quite fit the next agent's needs. The summary might be too long. The research list might miss a topic. The draft might be too formal.
This is normal. Don't add new tools — go back to the prompt that produced the bad output and tweak it. Add a sentence like "Keep each bullet under 20 words" or "Skip anything older than 7 days".
💬 Copy this prompt tweak: "From now on, only include items published in the last 7 days, and skip anything that isn't about AI tools."
You'll know it worked when the same agent, given the same input twice, produces noticeably better output the second time.
Common mistakes
- Giving one agent too many jobs. If your "Newsletter" chat is researching, summarizing, AND writing, the output will be confused. Split it into separate chats — one job per agent.
- Skipping the shared document. Without somewhere to pass work between agents, you'll end up copy-pasting huge chunks of text and losing track of what's what. Even a plain text file beats nothing.
- Trying to automate the handoffs on day one. Resist the urge to wire up fancy tools before you've manually run the team once. The hand-run version is where you learn what's actually broken.
- Writing prompts that are too vague. "Write a good summary" gives bad summaries. "Return 5 bullet points, each under 20 words" gives good ones. Always include the format and the length.