3 Claude Agents to Build First (Even If You Don't Code)

I used to spend 3 to 5 hours a week digging through YouTube to decide what to film. Now an agent does that job and I don't open a single tab. I have never written a line of code by hand, and the agent was built by describing the job in plain English. This guide shows the three agents I would build first, what each prompt is doing, and the order that gets you a working one fastest.

An agent is like a junior employee inside Claude. You hand it a job and it does the work without you holding its hand. It is not a robot and it is not a program you have to write. Under the hood it is a text file that holds three things: the agent's name, its one job, and the rules it follows.

That file is the agent. When you ask Claude Code to build one, Claude writes the file and saves it where you tell it to. If you can describe a job to a new hire, you can describe it to Claude Code. Most of the skill is deciding what the job is, not typing the instructions.

All three agents start the same way. You open Claude Code and type one short prompt in plain English. The prompt has three parts, and each one matters.

• A name. Pick something plain that says what the agent does. You will see this name in your files and when you schedule it, so make it obvious.
• One job. Say what the agent does in a single sentence. If you need the word "and" twice, you have two agents.
• Where to save it. Name the folder. I keep mine in a personal agents folder so I always know where they live.

In the video I show this with a throwaway example, a weekly recap agent whose one job is to write a short summary of the week. Claude writes the file with the name, the job, and the rules, and saves it. That is the whole mechanism. Every agent below is this same move with a different job.

Right now you probably check how your business is doing in five different places: your community, your email list, your ads, Instagram, and YouTube. The first agent puts all of it on one page, with a tab for each place you check.

Each tab shows the numbers you would otherwise open that app to find. Mine has tabs for Instagram and YouTube that I can show on screen, and tabs for my community and ads that I keep closed because they carry money numbers. Here is how it works under the hood: Claude Code wrote small programs, called scripts, that pull the numbers in. They refresh every 4 hours, so the dashboard keeps itself up to date.

Claude also does a few jobs on the dashboard itself. It pulls the email list numbers, and at the top of every tab sits a panel called this week's focus, where it says in plain words what to work on next.

What you type to get your own is a short description: one dashboard page, a tab for each place you check your numbers, your list of places, and how often it should refresh. The exact wording I use is inside the Claude Code Club.

Some people will tell you the dashboard is an automation, not an agent, and they have a point. It does not decide its own steps. It runs the same scripts on a timer, and Claude only steps in for a few of the jobs. So where is the line?

• It is an automation when the steps are the same every time and a timer runs them. The dashboard refresh is this.
• It is an agent when Claude has to read something and make a call, like picking which topic wins or deciding which email matters.
• Which one to pick: build whichever one does the job. Most people want the work done and do not care what it is called, so I keep both and call the whole set agents.

A dashboard only tells you where you are. It does not tell you what to do about it. The second agent answers that, and it is the one that sent me the message that opens the video. I call it my film brief because it briefs me on what to film.

Here is what it does. It is set to run at 7 every morning. It checks the 20 channels I follow and finds what jumped in the last 3 days. It checks my own titles so it does not repeat a video I have already made. Then it scores the topics and picks one.

I do not get a list. I get one video, with the title, the first lines to say, and what to show on screen. That is the job that used to take me 3 to 5 hours a week.

The most useful decision in the film brief is the output. A list of ten ideas puts the thinking back on you, which is the exact job you wanted to hand off. One pick with a reason hands off the thinking.

Here is a real result. One run told me the video to make was three Claude prompts for a $10,000 website. The reason it gave: a beginner website video from a channel with 349 subscribers was breaking out, and nobody had made the three-prompt version. That is a finding with a number and a gap, which is exactly what you need to decide in 10 seconds.

To build your own, describe the job the same way: a name, then every morning it checks a short list of channels, finds the video that jumped most in the last 3 days, checks that you have not made it already, and sends you one idea with a title and an opening line. Start with five channels, not twenty. The exact prompt is inside the Claude Code Club.

An agent you have to remember to start is a tool, not an employee. To make it run every morning, type /schedule in Claude Code, which is the command for running a job on a timer. It runs in the cloud, so your laptop can be closed.

That one detail changes what you can build. You are not leaving a computer on overnight. You describe the job, set the time, and the result is waiting in the place you check first. Mine lands in Slack.

The first two agents look at your business. The third looks at your day. It sends you a short email summary every morning, so you know what needs you today without opening your inbox.

Here is why it matters. A study of 40 office workers found they check their email 77 times a day on average. You open the tab to check one thing and you are back in it 10 times before lunch. The job is small: read my email for me and tell me what matters.

The prompt has four decisions inside it, and they are worth copying even without my exact words:

1. 1A window. Only unread email from the last 24 hours, so the summary stays short.
2. 2Three things to report. Who wrote, what needs a reply today, and which messages come from someone who might want to buy.
3. 3One summary. A single message, not a message per email.
4. 4A limit on what it may do. It can only read. It never sends, deletes, or edits anything.

The exact wording for the prompt is inside the Claude Code Club.

To let Claude see your email, you connect Gmail through the Claude app, on the connectors page. A connector is how Claude gets permission to use a tool like Gmail.

• Connect only Gmail for this agent and nothing else. Every extra connection is one more thing the agent can reach.
• Read-only is an instruction, not a lock. Telling the agent it can only read is a rule in your prompt. It is not a setting that physically blocks sending.
• So decide who writes the replies. I let this agent do the summaries and I write the replies myself.

Then type /schedule in Claude Code so it runs every morning before you sit down.

I build the inbox agent first, even though it is the least exciting of the three. The reason is speed of proof: you find out if it works by tomorrow morning.

• The dashboard takes a few tabs before it is useful.
• The film brief needs a list of channels and a rule for picking.
• The inbox agent needs one job and a simple connection.

The boring one also teaches the whole loop, and the other two run on that same loop. Describe a job. Connect one tool. Put it on a schedule. After you have done it once, the dashboard and the film brief are the same three moves with more pieces.

This is the idea that matters more than any of the three agents. Every agent in this guide is the same thing: a job you were doing by hand, written down, so Claude does it.

That changes how you pick what to build. You do not start with what is impressive. You start with the job you open a tab for 10 times a day. The dashboard replaces checking five apps. The film brief replaces hours of digging through YouTube. The inbox agent replaces the 77 checks a day. Each one began as something I was already doing by hand.

So the test for any new agent is short. Can you describe the job to a new hire in a few sentences? Do you know what a good result looks like? If yes, you can build it. If you cannot describe the job yet, do it by hand once more and write down what you did.

These agents are useful and they are not magic. Here are the limits I hit myself, and how to plan around them.

• A pull can fail. In the film brief run I show, one data pull failed and the agent had to run again. Build the habit of glancing at the result.
• The agent does one job. The film brief picks the video and does not film or edit it. The inbox agent summarizes and does not write replies for you.
• The dashboard is not deciding anything. It runs the same scripts on a timer. Treat it as a fast way to see numbers, not as an advisor.
• Money numbers need care. I keep my community and ads tabs closed when I share my screen because they show money numbers. Decide what you will show before you record or screenshare.

Here is the shortest honest path to a real result in one sitting:

1. 1Check that you are on a paid Claude plan, since Claude Code is not on the free plan.
2. 2Write down the job you open a tab for 10 times a day. Mine was email.
3. 3Open Claude Code and describe the agent in one short prompt: a name, one job, and where to save it.
4. 4Connect only the one tool it needs through the connectors page in the Claude app.
5. 5Tell it what it may do and what it may not do, in your own words, in the prompt.
6. 6Type /schedule so it runs every morning, in the cloud.
7. 7Read the first result tomorrow. If it works, build the next agent. If it does not, fix the one job before you add anything.

Build the boring one first. The exciting ones can wait. If you want the exact words for all three, plus the people building the same set, the Claude Code Club has them.