Live Masterclass: How to Build AI Agents with Claude Code

You have ideas. Dozens of them. But between having the idea and shipping it, there's a gap that used to require engineers, agencies, or months of learning to code.

That gap is closed.

In a live LinkedIn webinar with Ariel, I built three real AI agents from scratch - a company intelligence agent, a content writing agent, and a data dashboard - all without writing a single line of code myself. The entire audience watched it happen in real time.

This guide breaks down exactly what we built, the prompts we used, and how you can copy the same approach to build your own AI agents today.

Watch the Full Walkthrough

What You'll Build

Each of these was built live, on camera, using plain English prompts. No IDE. No coding bootcamp. Just Claude Code and clear instructions.

Key Concepts: Claude Chat vs Cowork vs Code

Before we build, you need to understand the three levels of Claude. Think of it as a housing metaphor:

• Claude Chat is a sandbox. Great for questions, brainstorming, analysis. But it can't touch your files or system.
• Claude Cowork is like an apartment. It can work on your computer - organize files, fill out forms, browse the web - but with guardrails.
• Claude Code is full access. It reads your files, writes code, runs terminal commands, and builds real applications. This is where agents live.

For everything in this guide, you need Claude Code (Max subscription, $100-200/month).

Part 1: Building the Company Intelligence Agent

This is the agent that changed how I prepare for every meeting, sales call, and partnership conversation. One command. Deep company intelligence. Every time.

What It Does

The /intel skill takes a company website and produces a structured intelligence profile:

• Value proposition - what they do, who they serve, how they position it
• Target audience & pain points - who they're selling to and what problems they solve
• Voice & tone - formality level, vocabulary, personality traits
• Competitive positioning - how they differentiate, key proof points
• Evidence - numbers, testimonials, results they cite

The Prompt

Here's the actual prompt we used to create the skill. You tell Claude Code to build it:

Create a Claude Code skill called /intel. It should analyze any company website I give it. Find and read the sitemap and key pages (homepage, pricing, about, features). Extract the value proposition, target audience and their pain points, and the company's voice and tone (formality, vocabulary, personality). Identify competitive positioning and key proof points (numbers, testimonials, results). Save the full analysis as a structured markdown file in profiles/[company-name].md. Make the output really structured and useful.

Running It Live

Once Claude creates the skill file, using it is this simple:

What the Output Looks Like

The generated profile is a structured markdown document - typically 2-4 pages - with sections you can immediately use:

This isn't a generic summary you'd get from ChatGPT. It's a structured, saved, reusable intelligence file that other agents can build on.

Part 2: Building the Content Writing Agent

Here's where things get powerful. The /write agent doesn't work in isolation - it reads the output from /intel.

How /write Depends on /intel

This is the key insight: agents build on each other's output.

When you run /write, it:

1. Reads the company profile generated by /intel
2. Asks what content type you want
3. Writes using the company's actual voice, vocabulary, and proof points
4. Follows anti-AI-smell rules so the output reads like a human wrote it

The Prompt

Create a Claude Code skill called /write. It should read the company profile from the profiles/ folder, then ask what type of content I want (LinkedIn post, blog outline, or email). Write the content using the company's voice, vocabulary, and proof points from the profile. Include anti-AI-smell rules - no 'delve', 'leverage', 'utilize', 'in today's landscape'. Use contractions. Mix sentence lengths. Be specific. After the content, add a brief note explaining the voice choices it made.

The Skill Ecosystem

The real power isn't any single agent. It's the system. Here's what we demonstrated in the webinar:

One research pass. Multiple content pieces. All personalized. All consistent. That's what an AI operating system looks like.

Part 3: Building a Dashboard

The final demo was the quickest - and blew people's minds the most. We asked Claude Code to build a visual dashboard that displays all the company profiles we'd generated.

No Figma. No React tutorial. Just:

Build me a simple web dashboard that reads all company profiles from the /profiles directory and shows each company as a card with key stats (name, industry, funding, team size). Let me click into any company to see the full profile. Clean, minimal design - dark sidebar, white content area. Run it locally so I can view it in my browser.

Claude Code read the profiles, built a Next.js app, styled it, and had it running on localhost in minutes. Live. On camera.

That's the gap closing in real time. The idea-to-built pipeline went from months to minutes.

The Bigger Picture: Your AI Operating System

What we showed in the webinar is just one use case. But the pattern scales to everything.

I'm not a developer. I'm what I call an idea engineer - someone who uses AI to turn ideas into reality without writing code.

The agents shown in this webinar are just the beginning. The same pattern works for:

• Meeting prep agents - auto-research everyone in your next meeting
• Email response agents - draft replies in your voice using conversation history
• Report generators - pull data from multiple sources into formatted reports
• Competitive analysis agents - monitor competitors and flag changes
• Social media agents - create platform-specific content from a single brief

Every one of these follows the same build pattern you just learned.

Principles That Make This Work

After building 30+ agents, these are the principles that separate agents that work from agents that don't:

1. Plan Mode First - Always

Never let Claude just start building. Enter Plan Mode (Shift+Tab twice), describe what you want, and review the plan before executing. This one habit will save you hours of backtracking.

2. Context Is King

The more context you give Claude, the better the output. That means:

• Reference specific files (read my brand.md for voice guidelines)
• Provide examples of what good output looks like
• Be explicit about format, length, and structure

3. Iterate, Don't One-Shot

Your first version won't be perfect. That's fine. Run the agent, review the output, refine the prompt, run again. Three iterations usually gets you to 90%+ quality.

4. Start Basic, Then Expand

Don't try to build a complex multi-step agent on day one. Start with a simple single-task agent. Get it working. Then add features, chaining, and edge cases.

5. Think in Systems

Individual agents are useful. Connected agents are transformative. Design your agents to produce output that other agents can consume. That's how you build an AI operating system, not just a collection of tools.

Your First Agent in 10 Minutes

Ready to build? Here's a copy-paste starter prompt. Open Claude Code and paste this:

That's it. Ten minutes from now, you'll have your first working AI agent.

From there, the playbook is simple: build one agent. Use it for a week. Notice what's missing. Build the next one. Before you know it, you'll have an AI operating system custom-built for how you work.

Don't want to start from scratch? The AI Builder Toolkit comes with 30+ pre-built skills, prompts, and workflows you can install in one click - including the /intel and /write agents from this walkthrough.

Want to build live with us? Join the next MasteringAI Workshop where we build agents together in real time, answer your questions, and help you ship your first skill before you leave.