Your Agent Control Center

Stop Vibing. Start Directing.

You’ve been training for this your entire life and you don’t even know it.

Every time you’ve explained what you need to a coworker, described a problem to a friend, or told someone exactly what you want for dinner… you’ve been practicing the single most important skill in the AI revolution.

Talking.

Not coding. Not “prompt engineering.” Not whatever LinkedIn is calling it this week. The ability to think about what you want, and then say it clearly. That’s it.

What Is an Agent Control Center?

Forget ChatGPT. Forget the chatbot in the corner of a website. Forget the AI that writes your emails weird. All of that is what I call vibing. You’re asking questions into a browser tab and hoping for something useful.

What we’re building is something different.

An Agent Control Center is three things working together: a free program called VS Code that lets you see and organize your files, an AI agent called Claude Code that understands plain English, and your actual files and folders on your actual computer.

You open VS Code. You see your files on the left side. And at the bottom, there’s a place where you just… type what you want. In English. “Create a project plan for my side business.” “Research the best way to soundproof a home office and put everything in a document I can keep.”

And then it does it. On your computer. In your files. Files you own, files you can open in any program, files that will still be there in ten years even if every AI company on the planet goes bankrupt tomorrow.

This is not a chatbot. This is a workspace.

The Spectrum

There are three levels to how people use AI right now, and most people are stuck on level one.

	Vibing	Vibe Coding	Directing
What it is	Chatting with AI in a browser	Letting AI write code while you watch	Directing AI agents from your desktop
Where it lives	Browser tab	Code editor	Your file system
What you get	Chat history (trapped in the platform)	Source code	Real files in your folders
When you close it	Gone	Code remains	Everything remains
Builds over time	No	No	Yes

Vibing is asking ChatGPT a question and getting an answer. It’s fine for quick stuff. But your conversation is locked inside someone else’s platform, and when you close the tab, it’s gone.

Vibe coding is aimed at building software. Not relevant if you’re trying to plan a course, organize a project, or learn something new.

Directing is what this guide teaches. You’re running an AI agent from your own computer. It reads your files, creates new ones, organizes your work, and remembers who you are. You’re not renting intelligence. You own the whole thing.

Two Things to Know Before We Start

AI is not the source of thought — it’s the accelerant. You bring the idea, the judgment, the direction. AI brings the speed and the ability to work through large amounts of information without getting tired. You think. It structures. You decide. It executes. You review. It revises. The thinking is still yours. It just happens faster.

The real skill isn’t asking better questions — it’s building the right environment. Most people think using AI well means crafting perfect prompts. It doesn’t. The real skill is building context around the AI so it already understands what you need before you say a word. Your files, your project structure, a simple document that tells the AI who you are — that’s what makes it work.

What You Need

A computer. Windows, Mac, or Linux. Desktop or laptop. It doesn’t need to be fancy. If it can run a web browser, it can run everything we’re about to install.

VS Code (free). This is the workspace. Think of it as a really smart file explorer with a built-in place to talk to AI. Programmers use it every day, but that doesn’t mean it’s only for programmers. It’s free, it works on every operating system, and millions of people use it.

A Claude subscription ($20/month). This is the AI agent. Claude is made by Anthropic, and the Pro subscription gives you access to Claude Code. Twenty bucks a month. No contract. Cancel any time.

That’s it. No special hardware. No additional software.

Step 1: Install VS Code

Go to code.visualstudio.com. Click the big download button. It detects your operating system automatically.

Windows: Run the installer. Click through the prompts. All the defaults are fine. When it asks about additional tasks, check the box that says “Add to PATH” and “Register Code as an editor for supported file types.” These make your life easier later.

Mac: Download the .zip file. Unzip it (double-click). Drag the VS Code icon into your Applications folder. The first time you open it, macOS might show a warning that says “VS Code is an app downloaded from the Internet. Are you sure you want to open it?” Click Open. This is normal — macOS warns you about every app that didn’t come from the App Store.

Linux: Go to code.visualstudio.com and download the .deb package (Ubuntu/Debian) or .rpm package (Fedora/Red Hat). Install it through your package manager or double-click the file. If you’re on Linux, you probably already know how to install software, so I’ll spare you the lecture.

Once it’s installed, open VS Code. You’ll see:

A sidebar on the left with icons
A big empty area in the middle
A welcome tab you can close

That’s your workspace. Don’t worry about the rest yet.

Step 2: Get Comfortable With VS Code

VS Code has a lot of buttons, but you only need to know about four things right now.

The File Explorer (top icon in the sidebar). This shows the files and folders in whatever project you have open. It works like the file explorer on your computer because that’s exactly what it is.

The Extensions panel (icon that looks like four squares). This is where you install add-ons that give VS Code new abilities. We’re about to use it.

The main editor area (the big space in the middle). This is where files open when you click on them. You can read and edit files here.

The “Open Folder” command. This is how you tell VS Code what to work on. Go to File > Open Folder and select any folder on your computer. VS Code now shows you everything in that folder. This is important: VS Code works with folders, not individual files. When you open a folder, the AI agent can see everything inside it.

The first time you open a folder, VS Code might ask “Do you trust the authors of the files in this folder?” Click “Yes, I trust the authors.” This is a security feature — VS Code wants to make sure you’re not opening someone else’s untrusted code. Since these are your own files, it’s safe.

Step 3: Install the Claude Code Extension

Click the Extensions icon in the sidebar (four squares). In the search bar at the top, type “Claude Code.”

You’ll see results appear. Look for the one that says “Claude Code” by Anthropic — that’s the official one. There may be other extensions with similar names. Make sure it says Anthropic as the publisher.

Click Install.

After a few seconds, you’ll notice a new icon in your sidebar — the Claude icon. You might also see a panel appear at the bottom or side of your screen. That’s the Claude Code panel. This is where you’ll type your directions to the AI agent.

Click the Claude icon in the sidebar to open the Claude Code panel. The first time, it will ask you to sign in.

If you already have a Claude Pro account: Click the sign-in button. It’ll open your web browser and ask you to log in to claude.ai. Log in, authorize the connection, and come back to VS Code. You’re connected.

If you don’t have an account yet: Go to claude.ai in your web browser. Sign up for an account and subscribe to the Pro plan ($20/month). Then come back to VS Code and sign in.

Important distinction: claude.ai (the website) is the chatbot in a browser. Claude Code (in VS Code) is the agent on your computer. Same company, same AI, completely different experience. The website is vibing. VS Code is directing. You’re paying for the Pro plan because it gives you access to Claude Code, which is what makes the Agent Control Center possible.

Once you’re signed in, the Claude Code panel should show a text input area. You’re ready.

Step 5: Your First Conversation

Let’s make sure everything works.

1. Create a folder. Open your file explorer (not VS Code — your regular file explorer). Create a new folder somewhere easy to find. Your Desktop or Documents folder is fine. Call it “My Agent Control Center” or “AI Projects” or whatever makes sense to you.

2. Open that folder in VS Code. Go to File > Open Folder and select the folder you just created. You should see an empty file explorer in the sidebar.

3. Open the Claude Code panel. Click the Claude icon in the sidebar. You should see a text box where you can type.

4. Type this exactly:

Hello, can you create a file called hello.md with a short introduction about what this workspace is for?

Press Enter (or click the send button).

5. Watch what happens. The agent will respond in the panel. It will tell you what it’s about to do. It might ask for permission to create a file — click Allow or Yes. And then… a file appears in your sidebar. hello.md. Click on it. There’s your content, sitting in a real file on your real computer.

What you might see that’s unfamiliar: The agent might show you a “diff” — a view that highlights what it’s adding to a file in green. This is normal. It’s showing you the changes before (or after) it makes them. You might also see text scrolling in a terminal panel at the bottom of the screen. That’s the agent working. You don’t need to understand it.

What the permission prompts mean: Claude Code will sometimes ask “Allow Claude to create this file?” or “Allow Claude to run this command?” This is a safety feature. The agent is asking before it touches your computer. For now, click Allow. As you get comfortable, you’ll develop a sense for when to allow and when to pause and read what it’s proposing.

You just directed an AI agent from your desktop. That file isn’t trapped in a chat log. It’s not locked inside someone else’s platform. It’s a plain text file in a folder you control.

If anything didn’t work, check the Troubleshooting section at the end of this guide.

Pick Something Real

That hello file proved everything works. But it wasn’t real work. You don’t need an AI agent to write you a greeting.

This is the single most important instruction in the entire guide. Do not skip it.

Pick something you actually need to get done. Not a hypothetical exercise. Something that’s been sitting in the back of your mind — a plan you haven’t made, a letter you haven’t written, research you’ve been putting off.

A few examples:

A business plan for a side project you’ve been thinking about
A research document about a topic you need to understand for work
A lesson plan if you teach anything
A budget for an upcoming trip or event
A comparison of options you’re deciding between
A guide you wish existed for something you know about

Pick one. Something specific. Something where you’ll actually use the output.

Start Talking

Open your project folder in VS Code. Open the Claude Code panel and describe your project. Not in keywords. Talk to it the way you’d talk to a smart assistant on their first day.

“I’m planning a two-week trip to Portugal in September for two people. We want to split time between Lisbon and Porto, with maybe a day trip to Sintra. Budget is around $5,000 total not including flights. Can you create a rough itinerary, a packing list, and a budget breakdown? Put each one in its own file.”

Or:

“I need to understand how property taxes work in New Jersey. I just bought my first house and I have no idea what to expect. Can you research this and create a plain-English guide that covers what I’ll owe, when it’s due, how assessments work, and what I can do if I think the assessment is wrong?”

Notice what’s happening. You’re not writing a “prompt.” You’re not trying to hack the AI with magic words. You’re just explaining what you need.

Direct the Revision

Files appear in your sidebar. Click on one. Read it.

Some of it will be exactly what you wanted. Some of it won’t be quite right. That’s fine. This is a first draft, not a final product. You’re not done — this isn’t vibing. You don’t just accept whatever came out.

Find something that’s off. Tell the agent what to change. Be specific.

“The budget is missing food costs. Add a section for meals, assuming we eat out twice a day at mid-range restaurants.”

“This packing list has way too many clothes. We’re packing carry-on only. Revise for a minimalist approach.”

The agent updates the actual file on your computer. You can see the changes right there.

This is the workflow. You direct. The agent executes. You review. You redirect. The files get better with each pass. Every revision builds on what’s already there — the agent reads the files it created. It doesn’t start from scratch.

It Reads What’s Already There

This is the part that surprises people. When you open a folder in VS Code, the agent can read everything in it. Every file. Every subfolder.

If you already have notes, documents, half-finished plans — the agent can see all of it. You don’t have to copy and paste anything. You don’t have to re-explain what exists. You point it at a folder and say “look at what’s here and help me with it.”

Most AI tools start from blank. Your Agent Control Center starts from what you’ve already got.

And the same workspace where you plan is the same workspace where you execute. No switching tools. No “I’ll get to that later.” The plan file and the project files live side by side, and the agent knows about all of them.

Teaching It Who You Are

So far, every conversation starts from zero. The agent doesn’t know your name, what you’re working on, or how you like things done.

That’s about to change.

There’s a file you can create called CLAUDE.md. It’s a plain text file that sits in your project folder. The AI agent reads this file automatically at the start of every conversation. Before you say a word, it already knows whatever you put in there.

Ask the agent:

“Create a file called CLAUDE.md. I’m going to tell you about myself and this project, and I want you to put it in that file.”

Then talk. Tell it:

Who you are. Your name, what you do, what you’re working on.
What this project is. What you’re trying to accomplish. Who it’s for.
How you like to work. Short answers or detailed? Bullet points or paragraphs? Formal or casual?
What to avoid. Fluff, disclaimers, overly formal language, emojis, whatever annoys you.

Here’s an example:

# About Me

I'm Sarah. I'm a high school biology teacher in Austin, Texas.
I'm using this workspace to build supplemental lesson materials
for my AP Bio classes.

# This Project

Creating study guides for Unit 4 (Cell Communication and Cell Cycle).
Each guide should be 2-3 pages, written for 17-year-olds who are
smart but might not love biology yet.

# How I Work

- Keep answers concise. I'd rather ask a follow-up than wade
  through a wall of text.
- Use tables and bullet points for comparisons.
- Don't hedge. State facts as facts.
- Don't add filler like "Great question!" Just answer.

That’s it. A short document that tells the AI who you are and what you need.

Shortcut: If writing it from scratch feels like too much, try this after your first project session: “Create a CLAUDE.md file that captures everything you’ve learned about me, this project, and how I like to work based on our conversation so far.” The agent writes the first draft from what it observed. You review and refine.

Why This Changes Everything

Next time you open this workspace, the agent already knows you’re Sarah, you teach AP Bio, you’re building study guides, and you want concise answers without fluff. You walk in and say “write the next study guide on signal transduction pathways” and it knows exactly what that means, who it’s for, and how you want it written.

No preamble. No re-explaining. You pick up right where you left off.

And it gets better over time. As you notice things, you update the CLAUDE.md. “Always include a 5-question quiz at the end of each study guide.” Done. Every future study guide gets a quiz.

This is the concept that separates an Agent Control Center from every chatbot you’ve ever used. Chatbots are stateless — every conversation is isolated. Your Agent Control Center is stateful. Your files are there. Your CLAUDE.md is there. The work you did last week is there. The workspace gets smarter, not because the AI is getting smarter, but because the context around it is growing.

Here’s what makes that context powerful: the more specific you are, the better the output. A CLAUDE.md that says “I’m a teacher” gets generic results. One that says “I teach AP Bio to 17-year-olds who are smart but bored, and I need 2-3 page study guides with quizzes” gets results you can actually use. Specificity is the single highest-leverage thing you can do.

A few tips:

Start small. A few lines about who you are and what the project is. Add more as you discover what the agent needs to know.
Be specific about what annoys you. “Don’t use emojis.” “Don’t start with ‘Great question.’” “Don’t add disclaimers.” Small instructions, huge impact.
Update it. If the project shifts or you want a different format, change the file. Ten seconds. Saves you from repeating yourself forever.
One per project. Different folders can have different CLAUDE.md files. Your trip folder has one about the trip. Your business folder has one about the business. Each workspace is customized.

How to Learn Anything

You’ve got a working Agent Control Center. Now let’s use it.

You want to understand something new. How the Federal Reserve works. How solar panels generate electricity. How to read a financial statement. Whatever it is.

Here’s the workflow.

Start broad. Tell the agent to create a structured overview. Not a chat answer — a document. In files.

“I want to understand how the Federal Reserve works. Create a structured guide that covers: what the Fed actually is, how it controls interest rates, what quantitative easing means in plain English, and why any of this matters to a regular person. Assume I’m smart but have zero economics background. Put each major topic in its own file.”

A folder of organized files appears. Each one covering a specific topic. Written at your level. Saved on your computer.

Go deep where you’re confused. Read the overview. Find the parts that don’t click. Direct the agent to expand on those specific areas.

“In the interest rates file, the section on federal funds rate vs. prime rate is confusing. Rewrite it with a concrete example — trace what happens when the Fed raises rates by 0.25% through to what a regular person pays on a car loan.”

The agent reads the file it already wrote, finds the section, and rewrites it. Your learning resource just got better. And it stays better because it’s a file, not a chat message that scrolled past.

Connect it to what you know. Tell the agent about your background. “I’m an electrician, so I understand circuits. Explain solar cells using electrical concepts I already know.” The explanation gets tailored to you.

Test yourself. Ask the agent to create a quiz based on your files. See if you can answer without looking. If you can’t, you know where to go deeper.

Save and return. The files are there tomorrow. Next week. Next year. You don’t re-learn what you’ve already learned because it’s saved, organized, and waiting.

How to Do Anything

Same workspace, different goal. You don’t want to learn about something. You want to get something done.

The pattern is the same: describe what you need, get a structure, execute within it, revise as you go.

“I want to start a YouTube channel about woodworking. Help me think through what I’d need. Not a generic checklist — walk me through the actual decisions: what kind of videos, what equipment under $500, how to structure the first three episodes, and a realistic posting schedule for someone with a full-time job.”

Files appear. A plan with specifics. Episode outlines. An equipment list. A schedule.

Now pick one thing from that plan and do it in the same conversation.

“Let’s start on Episode 1. The outline says it’s an introduction video. Write a rough script — casual tone, about 5 minutes spoken, include notes about what to show on camera.”

A script file appears. Some parts sound like you, some don’t. Direct revisions. The script gets better. You’re executing the plan in the same workspace where you made it.

This is the gap closer. Not better plans — better follow-through. The same place where you think is the same place where you do.

Three Practices That Make This Work

These come from hundreds of hours of working this way. They’re the difference between getting decent results and getting results you can actually use.

Verify what you get. The agent will tell you “Done!” and what it produced will look confident and polished. Don’t just accept it. Read it. Check it. If the agent wrote a guide about property taxes, spot-check a few facts. If it created a budget, run the numbers yourself. AI output that looks right but isn’t is worse than no output at all — because you’ll trust it. The rule: if something matters, verify it before you use it.

A useful trick: ask the agent “What are you least confident about in this document?” It will often flag the exact parts that need human verification.

One thing at a time. Don’t ask the agent to plan your business, write your website copy, and organize your finances in one conversation. Big vague requests get big vague results. Instead, break it down. One project. One task. Check the output. Then move to the next thing. Small, verified steps compound faster than ambitious leaps.

Think before you ask. Before you type a direction, spend ten seconds thinking: what do I actually need? What does “done” look like? What constraints matter? The more you can answer these before you start, the less back-and-forth you’ll need. You’re not asking a search engine for keywords. You’re briefing a team member. The clearer your brief, the better the first draft.

What This Is Not

Your Agent Control Center is a starting line, not a finish line. It gets you oriented fast, builds structure from vague ideas, and handles the tedious parts. But it can’t replace deep understanding. Reading a book, taking a hands-on class, struggling with a problem until you figure it out — those things still matter.

What it can do is get you to the starting line faster. Before you dive into a topic, you can build a map of the territory. Before you start a project, you can have a plan that’s specific to your situation. Before you write something, you can have a draft to react to instead of a blank page.

AI is a multiplier. What it multiplies depends on what you bring to it.

What’s Next

You’ve got the tools. You’ve got the principles. You’ve done real work with your Agent Control Center.

There’s a lot more this setup can do — memory systems that persist across sessions, commands that trigger complex workflows with a single word, rules that enforce your preferences automatically, and ways to connect your workspace to the outside world. That’s Part 2.

But you don’t need any of that to start getting value right now. You already have everything you need. A workspace. An agent. Your files. Your direction.

Stop vibing. Start directing.