How to build a Claude Code AI content agent in 20 minutes (no coding)

31 Oct 2026 04:07 21,158 views

Learn how to turn Claude Code into a fully working AI content agent that plans, designs, and animates social posts for you—without writing a single line of code. This step-by-step guide covers setup, skills, memory, and running your agent in the cloud.

AI chatbots are great at answering questions, but AI agents actually do the work for you. With Claude Code, you can now build your own agents that plan, create, and even animate your content—without writing a single line of code.

This guide walks you through how to build a Claude Code content agent that turns your videos and newsletters into on-brand, animated Instagram carousels. You’ll go from a blank workspace to a self-improving, cloud-based agent that runs on its own.

What makes an AI agent different from a chatbot?

A chatbot responds to whatever you type. An AI agent takes a goal, makes decisions, and uses tools to get the job done for you.

A useful way to think about agents is as a new teammate. You don’t just give them answers—you give them responsibilities and let them decide what to do first, which tools to use, and when a result is good enough.

In this setup, every agent is built from five core parts:

1. Brain: The AI model that does the thinking (for example, Claude 3.5 Opus in Claude Code).

2. Context: What the agent knows about your brand, content, audience, and style.

3. Job: The specific goal it’s responsible for, like “turn long-form content into Instagram carousels.”

4. Rules: How it should and shouldn’t work—your guardrails, preferences, and constraints.

5. Tools: What it’s allowed to use: files, websites, Notion, GitHub, and more.

Once those five pieces are in place, you have the foundation of a real agent, not just another chat.

Level 0: Set up your agent workspace

Before you “hire” an agent, you need to prepare its office. In practice, that office is just a folder on your computer that Claude Code can work inside.

For example, you might create a folder called Business OS. This becomes the central place where all your agents, files, and instructions live.

Inside that folder, add a few key items your content agent should know about:

Your logo and brand assets
A few personal or brand photos
Examples of real posts you like (e.g., Instagram carousels you’ve already published)

Don’t overthink it. A handful of strong, on-brand examples is enough to get started.

Next, open the Claude desktop app and switch to the Code tab. Point Claude Code at your new Business OS folder so it can see the files you just added.

Choose how Claude Code runs and which model to use

Claude Code can run in two places:

Local: Everything runs on your computer and uses your local files.
Cloud: Everything runs remotely so your agents can keep working even when your laptop is closed.

Start with local for fast iteration. You can move to the cloud later.

Then set:

Permission mode: Use Auto so the agent can work without constantly asking you, while still blocking risky actions.
Model: Pick a strong reasoning and instruction-following model like Claude 3.5 Opus, especially for content, branding, and design-heavy tasks.

Onboard Claude with your brand context

Every time you start a new Claude Code session, it begins with a blank mind. It only knows what you tell it or what it reads from your files. That’s why onboarding matters.

Your first job is to have Claude create two key files in your workspace:

CLAUDE.md: A simple onboarding guide that Claude reads at the start of every session.
Brand context file: A document that captures your business, audience, voice, visual style, and examples.

You can ask Claude in plain language to:

“Read my website, the files in this folder, and create: (1) an onboarding guide called CLAUDE.md, and (2) a brand context file that covers my business, audience, voice, and visual style.”

Claude will fetch your website, scan your example posts and brand files, and ask you clarifying questions such as:

Which fonts do you use?
What do you like about your existing posts?
What colors and layout styles feel ‘on-brand’ to you?

After a short back-and-forth, you’ll see:

CLAUDE.md at the root of your workspace
A brand folder containing your brand context file

To confirm it’s working, start a new Claude Code session and ask for a simple test, like a weekly content plan. You’ll see that Claude now pulls from your brand context instead of making generic suggestions.

If you want an even faster path from zero to a working agent, you may also like this short guide to building your first no-code AI agent.

Level 1: Build your first Claude Code content agent

With your workspace ready, it’s time to turn Claude into a dedicated content agent.

The easiest way is to use a prewritten “agent designer” prompt. The idea is simple:

You paste in the agent designer prompt.
You describe the job you want the agent to do.
Claude interviews you and then designs the full agent configuration for you.

In this example, the job is:

“Turn my long-form content (videos and newsletters) into Instagram carousel posts, with up to 10 slides, in my brand style.”

Claude will ask clarifying questions about:

What platforms you care about (Instagram, LinkedIn, etc.)
Preferred angles, hooks, and tone of voice
Design preferences for covers, inner slides, and CTAs

Then it will summarize the agent using the five-part framework:

Brain: Which Claude model to use
Context: Which brand files and guides to read
Job: The exact responsibility of the agent
Rules: What to always do and what to avoid
Tools: Files, web access, and connectors it can use

Claude will also create a new folder and an agent file, for example content_creator_agent, inside your workspace. That file is the definition of your agent.

Using your content agent inside Claude Code

Once the agent is created, you don’t have to touch any configuration files. You can just tag it in chat.

Type @ and select your content creator agent, then give it a job. For example:

“@content_creator_agent Here’s a video transcript. Turn it into a 10-slide Instagram carousel in my style.”

The agent will then:

Read your transcript
Consult your brand context and examples
Propose angles and hooks
Generate a first-draft carousel layout and content

The first draft may not be perfect. You might notice issues like missing images, basic layouts, or font sizes that don’t feel right. That’s normal. At this stage, your goal is to get something usable and then improve it over time.

Level 2: Make your agent reliable with skills

Right now, your agent can create good posts, but each run might feel a bit different. Maybe the hook style changes, or the layouts swing from minimal to busy. To get consistent results, you need agent skills.

Think of skills as your agent’s playbooks—step-by-step instructions for how to perform a specific type of task every time.

For a content agent, a core skill might be a content playbook that covers:

How to structure a carousel (cover, story flow, CTA)
How to translate long-form content into short, punchy slides
How to handle motion and animation for each slide
Design rules: spacing, fonts, colors, icon style

A simple way to create this is to ask Claude to interview you about how to improve the last carousel it made. For example:

“Interview me to design a content playbook skill for my content creator agent, based on the carousel you just created. Think hard and ask detailed questions.”

Claude will ask deeper questions about:

The emotional energy your brand should have
Your signature style of hooks and headlines
How much motion and what types of animations you like
What makes a slide feel ‘finished’ to you

After you answer, Claude will generate a content playbook skill file inside your Claude folder. This file encodes the workflow and checklist your agent will follow.

If you want a more detailed breakdown of skills, check out this guide to AI agent skills and how to build one in minutes.

Testing skills with commands and files

Claude Code lets you trigger skills using /slash commands. When you type /, you’ll see a list of available skills, including the content playbook you just created.

For example, you might run:

“/content_playbook Take the carousel we just made (see attached file) and animate it.”

You can tag any file in your workspace using @filename, which makes it easy to reference specific carousels, transcripts, or guides.

The result: Claude takes your static carousel and generates animated versions of each slide—drawing diagrams, revealing text, animating shapes, and exporting every slide as an individual video file ready for social media.

Once you like how the skill behaves, assign it to your content agent so the agent automatically uses that playbook whenever it creates or animates posts.

Level 3: Give your agent memory so it improves over time

Even with a solid playbook, you’ll still give feedback: a color feels off, a background is too repetitive, an icon doesn’t match your brand. If the agent forgets this every time, you’ll keep repeating yourself.

The solution is to give your agent memory.

In this setup, your agent now reads three things before every job:

Onboarding guide (CLAUDE.md): What it learns on day one.
Skills (playbooks): How to perform specific tasks.
Memory file: Lessons from your past feedback.

Ask Claude to create a memory file for your content agent, for example:

“Create a memory file for my content creator agent, store it next to CLAUDE.md, and read it before every job. Only save feedback that applies to future posts, not one-off edits.”

Claude will generate something like content_creator_agent.memory.md and wire it into your agent’s workflow.

Now, when you review a carousel and say:

“No brown boxes—this color isn’t in my brand palette.”
“Inner slides should use different background colors for variety.”
“Replace this robot icon with my usual character.”

Claude will both:

Fix the current carousel
Summarize the reusable parts of your feedback into the memory file

On the next project, the agent reads that memory and avoids repeating the same mistakes. Over time, your agent’s judgment gets closer and closer to how you’d design things yourself.

Level 4: Move your agent to the cloud

So far, everything runs on your laptop. That’s fine when you’re actively working, but your agent stops when your computer is closed.

To let it keep working while you’re away, you need to move the system to the cloud. Claude Code can now run agents in the cloud using a GitHub repository as the source of truth for your files.

Step 1: Add external tools like Notion

First, connect any external tools your agent should use. For a content agent, Notion is a great place to store drafts and tasks.

In Claude:

Go to Customize → Connectors.
Connect your Notion account (or any other tool you prefer).
Tell your agent it’s now allowed to use Notion as a tool.

From now on, your content agent can read tasks from a Notion “Content Inbox” database and write drafts back into Notion for review.

Step 2: Sync your Business OS folder to GitHub

Next, create a private GitHub repository that mirrors your Business OS folder. On GitHub, a project folder is called a repository (or repo).

If you’ve never used GitHub before, you can create a free account and let Claude handle most of the setup. Ask Claude Code to:

“Sync a copy of my Business OS folder to GitHub and create a private repository called Business-OS on my account.”

Claude will guide you through logging in and push your files to GitHub. That repo becomes the cloud version of your agent workspace.

Step 3: Run your agent from the Claude mobile app

Once your repo is set up, you can access your agent from anywhere using the Claude mobile app.

From your phone:

Open Claude and go to the Code tab.
Create a new session linked to your GitHub repository.
Ask your agent to check your Notion content inbox and turn any pending ideas into carousels.

The agent will run as a background task in the cloud. You can lock your phone and it will keep working. When it’s done, it will reply with links to the new drafts in Notion, ready for your review.

Automate everything with scheduled tasks

At this point, the only thing left is that you still have to ask the agent to run. The final step is to set up a scheduled task (a routine your agent runs automatically on a schedule).

In Claude Code, go to the Schedule section and create a new task, for example:

Name: Daily Content Drop
Schedule: Every evening
Instructions: Check Notion’s Content Inbox for new items, turn each into an animated carousel using the content playbook skill, write the drafts back to Notion, and mark the tasks as done.

Under Advanced settings, be sure to select your GitHub repository so the task runs in the cloud against your Business OS files.

Optionally, you can even tell the scheduled task to:

“Propose improvements to the content playbook based on my feedback and update the system to keep improving.”

Now, whenever you drop ideas into your Notion inbox during the day, you’ll wake up to ready-to-review animated carousels the next morning—no manual prompting required.

How to apply this beyond content

The process you’ve just seen isn’t limited to Instagram posts. You can reuse the same four levels for almost any type of AI agent:

Research and summarization agents
Customer support drafting agents
Sales outreach and follow-up agents
Internal documentation or SOP agents

The pattern is always the same:

Set up a workspace with your files, examples, and brand context.
Build an agent with a clear job, tools, and rules.
Create skills so it follows consistent playbooks.
Add memory so it learns from your feedback.
Move it to the cloud and use schedules so it runs without you.

You don’t have to do all four levels at once. Start with a simple local agent that helps with part of your workflow, then layer on skills, memory, and automation as you go.

Next steps

If you’re using Claude Code (especially on a higher-tier plan), you already have everything you need to build practical, no-code agents that feel like real teammates.

This week, you could: