Claude Code Mastery: Build & Sell AI Workflows
Summary of the video “Build & Sell with Claude Code (10+ Hour Course)” by Nate Herk | AI Automation.
A comprehensive 10-hour course teaching you to build agentic AI workflows with Claude Code, from zero to pro. Learn the WAT framework (workflows, agents, tools), deploy automations that run 24/7, and monetize your skills by selling AI solutions to businesses at premium prices using value-based pricing.
Why Agentic AI Matters Now
The Agentic AI Market Explosion
The agentic AI market is projected to grow from $8 billion today to $40-50 billion by 2030, with 25% of enterprises already deploying agentic pilots this year and 50% expected by 2027. This represents a fundamental shift in how businesses automate work, creating massive opportunities for people who can build these systems.
Enterprise Adoption Timeline
Enterprise adoption of agentic workflows is accelerating rapidly. Currently 25% of major companies are running agentic pilots; this will jump to 50% by 2027, meaning half of Fortune 500-type businesses will rely on these systems within two years.
Why Traditional Automation Is Hitting a Ceiling
Traditional automation tools like Zapier and Make require you to manually map every step, handle edge cases yourself, and fix things when they break. Agentic workflows flip this: you describe the outcome, and the AI agent figures out the steps, handles unexpected situations, and adapts in real time during the build process.
Self-Healing vs. Deterministic Trade-off
While working with Claude Code interactively, the agent can self-heal and adapt mid-run if something breaks. However, once deployed to run automatically on a schedule or webhook, you're deploying the code and tools, not the agent itself—so it behaves like traditional deterministic automation. The real advantage is that the agent handled edge cases during the build, resulting in fewer bugs and better outcomes before deployment.
Getting Started with Claude Code
Five Ways to Run Claude Code
Claude Code can run in the terminal (most power, steepest learning curve), desktop app (visual, easiest for non-developers), web browser (no local setup, runs in cloud), IDE extensions like VS Code (zero context switching, see files and code together), or on a VPS (always-on, direct access to infrastructure). Each has trade-offs between power, ease, and customization.
Pricing and ROI of Claude Code
Claude Code requires a paid Claude subscription starting at $17/month (Pro) or $200/month (Max). The Max plan is recommended because a full-time software engineer costs ~$11,000/month in salary, so paying $200/month for AI-powered development is transformational ROI. Anthropic itself built Claude Co-work in 10 days with 2-3 developers using Claude Code—work that would normally take months.
Understanding Tokens and Context Windows
Tokens are how AI reads text—roughly 3-4 characters per token or 75% of a word. Claude's context window (working memory) is ~200,000 tokens. Everything you send (system prompt, files, conversation history, tools) consumes tokens. When the window fills up, quality drops (context rot), you hit usage limits, and costs increase. Commands like /context show usage, /compact compresses history, and /clear starts fresh.
The Cloud.md System Prompt
Cloud.md is a markdown file that acts as a system prompt for your entire project. Every time you send Claude Code a message, it reads cloud.md first. It should contain what you're building (purpose), why (goals), and how (frameworks, folder structure, key packages). Keep it lean to minimize token usage. Use /init to auto-generate one from your codebase.
Permission Modes: Plan, Ask, Bypass
Plan mode lets Claude think and ask questions without making changes. Ask before edits requires approval for file changes but auto-approves bash commands. Bypass permissions gives full autonomy. Use plan mode to nail requirements before building, then switch to bypass once you're confident in the direction.
The WAT Framework: Workflows, Agents, Tools
WAT Framework Overview
WAT stands for Workflows (instructions in markdown), Agents (the AI coordinator that reads workflows and decides which tools to use), and Tools (Python scripts that do the actual work). Workflows are like job descriptions or SOPs. The agent is the project manager. Tools are the workers. This structure keeps projects organized and lets Claude Code improve over time as it learns from outputs.
Workflows as Recipes
A workflow is a markdown file describing a process step-by-step, like a recipe. Example: a newsletter workflow might say 'research topic, generate HTML, create infographics, send via email.' The workflow doesn't execute these steps itself; it tells the agent what to do, and the agent figures out which tools to call and in what order.
Tools as Reusable Components
Tools are Python scripts that do one specific job: scrape a website, generate a PDF, send an email, analyze data. They're modular and reusable across workflows. Claude Code auto-builds and auto-fixes tools when they fail. You don't write the code; Claude Code does, and it updates tools based on errors it encounters.
Building Your First Agentic Workflow: Newsletter Example
Planning Before Building
Use plan mode to clarify requirements before Claude Code starts building. Ask ambiguous questions and let Claude ask you clarifying questions about tech stack, delivery method, brand assets, and edge cases you haven't considered. This ensures you're aligned before any code is written.
Newsletter Workflow Tech Stack
For the live newsletter demo, the stack was: Perplexity for research, Claude for content writing, Nano Banana for AI-generated infographics, HTML for email formatting, and Gmail for delivery. Brand assets (logo, guidelines) were provided so newsletters stayed on-brand. This shows how Claude Code orchestrates multiple APIs and services.
Self-Healing During Development
When the newsletter workflow hit a Unicode encoding error and an incorrect API endpoint, Claude Code diagnosed the problem, researched the fix, updated the tools, and retried automatically. This is the power of agentic workflows during development—the agent catches and fixes issues without you having to debug manually.
Human Review Checkpoints
The workflow included a human review step where Claude Code asked for approval of the subject line before sending. This lets you maintain control over critical outputs while automating the rest. You can approve, reject, or ask for changes using natural language.
Iterating and Improving
After the first newsletter run, the HTML formatting was broken. Instead of manually fixing the code, you just told Claude Code 'this is horrible, fix it.' It diagnosed the issue, updated the workflow and tools, and sent a better version. Each iteration improves the system because Claude Code learns and updates its own instructions.
Deploying Workflows to Production
From Interactive to Automated
Once you trust a workflow built in Claude Code, you export the code and deploy it to infrastructure like Trigger.dev, Modal, or Vercel so it runs automatically on a schedule or webhook. The self-healing agent is gone (you're deploying code, not the agent), but the workflow is battle-tested and deterministic, so it runs reliably.
Skills: Reusable System Prompts
Skills are system prompts you can load into Claude Code to improve specific tasks. Example: a front-end design skill makes Claude Code much better at building websites. You can create custom skills (e.g., 'make infographics look polished with our logo') so every time that task runs, it reads the skill first and produces consistent, high-quality output.
Selling AI Workflows: Positioning & Pricing
The Doctor vs. Pharmacist Positioning
Most beginners position themselves as pharmacists—they just fill prescriptions (build what clients ask for). Winners position as doctors—they diagnose the real problem, ask questions, identify constraints, and then prescribe the solution. Businesses pay premium prices for diagnosis and strategy, not just execution. Lead with 'I can save you X hours/month' not 'I build AI workflows.'
Value-Based Pricing Framework
Don't charge hourly or by complexity. Charge based on business impact: time saved, money saved, errors eliminated. If a workflow saves a business $12,000/year, charge 10% of that ($1,200) as a starting point so they see 10x ROI in year one. This anchors price on real metrics, not your effort, and builds trust because you can explain the math.
Real Example: Sales Process Automation
A client spent 1 hour/day qualifying leads at $40/hour = $800/week = $3,200/month = $38,400/year. A workflow automating this was priced at $5,500 (15% of annual savings), plus $550/month maintenance (10% of project cost). The client saw ROI in 2 weeks and got ongoing support. This is how you position and price professionally.
Retainers vs. One-Off Projects
Start with value-based project pricing to build trust. Once you deliver results, move to monthly retainers ($1,500-$15,000/month) for maintenance, optimization, and expansion. Retainers are better than one-off projects because you get stable income, deeper relationships, and more opportunities to expand. Protect 50-70% margins on retainers to cover your costs and team.
The 7-Day Client Acquisition Framework
Day 1: Set loose direction (not hyper-specific niche). Days 2-3: Have 5-10 warm conversations to understand pain points. Days 4-5: Propose free pilot on clearest pain point. Days 5-6: Build small MVP. Day 7: Discuss maintenance or expansion options. This loop repeats—each cycle gives you more proof, confidence, and positioning before cold outreach.
Start Warm, Not Cold
Begin with warm outreach (friends, family, past colleagues, community connections) before cold email. Warm leads convert faster and feel less awkward. Build a trust map of 20 people you know, have low-pressure conversations, and offer free pilots. Only move to cold outreach after you have 2-3 successful projects and proof.
Handling Objections Without Discounting
When clients push back on price, adjust scope, not price. Remove features, break into phases, reduce complexity. Never discount the value. If they're budget-sensitive, remind them of long-term ROI. If they undervalue your work or seem difficult early on, walk away—protect your energy and reputation.
Delivering & Maintaining Workflows
Hosting Options: Client-Hosted vs. Self-Hosted
Best practice: clients host their own Naden instance (or cloud equivalent) and invite you as a builder. This keeps you compliant, gives clients control, and avoids marking up hosting costs. Alternative: you host on your own infrastructure for internal automations only. Never host as a SaaS product without a commercial license.
API Key Management: Client Ownership
Clients should own and pay for their own API keys and usage. They sign up for the tool, generate the key, and paste it into the workflow. This keeps costs transparent, predictable, and avoids confusion. If they're intimidated, send a Loom video showing exactly where to click. Never mark up their usage or hide billing.
Security & Data Privacy
Credentials in Naden are encrypted at rest and decrypted only when the workflow runs. Use HTTPS for webhooks, sign secrets for verification, and never put sensitive data in URLs. For GDPR/regulated data, use data minimization (only pull needed fields), limit log access, and ensure clients can honor data deletion requests. Self-hosted Naden gives true data sovereignty.
Testing & QA Before Handover
Get real sample data from the client before testing. Plan for failure: test edge cases, bad data, duplicates, timeouts. Build guardrails so failures break safely and log errors. Run internal QA for days, then client-facing QA. For AI outputs, check relevance, tone, consistency, and safety. Log everything so you can show the client what you tested and why.
Handover Deliverables
Duplicate the workflow (keep test version, deploy clean production version). Back up JSON exports to GitHub or Google Drive. Clean up naming and add sticky notes explaining logic. Record a 1-2 minute Loom walkthrough showing how it works. Document what's included and what counts as 'done.' Make it so anyone on their team can understand the workflow later.
Maintenance Retainers & Recurring Revenue
After project completion, offer maintenance ($200-$1,500/month) to keep workflows healthy as APIs change and models update. Also offer optimization (weekly/monthly reviews) and expansion (V2 features). Price retainers as 10-25% of original project cost or as hourly packages. Recurring revenue makes income predictable and deepens client relationships.
Legal & Scope Documentation
Put scope, definition of done, payment terms, maintenance, service levels, bug vs. feature requests, IP ownership, and exit terms in a written agreement. This prevents scope creep, clarifies what 'finished' means, and sets expectations. When both sides know what they're buying, projects run smoother.
Advanced Topics & Monetization
Sub-Agents & Agent Teams
You can build multiple agents that work together or delegate to each other. Sub-agents can be lighter-weight (Haiku model) for specific tasks, while the main agent (Opus) handles complex reasoning. Agent teams let you parallelize work and scale beyond what a single agent can do.
Browser Automation with Claude Code
Claude Code can now control a browser: open tabs, take screenshots, click buttons, fill forms, scroll, and read page content. This unlocks workflows that interact with web apps that don't have APIs, like legacy systems or SaaS tools without integrations.
MCP Servers & External Integrations
MCP (Model Context Protocol) servers let Claude Code connect to external tools and data sources. You can build custom MCPs for proprietary systems, databases, or specialized APIs. This extends Claude Code's reach beyond standard integrations.
RAG: Retrieval-Augmented Generation
RAG lets Claude Code pull relevant context from documents, databases, or knowledge bases before generating responses. This makes outputs more accurate and grounded in real data. Useful for customer support bots, research workflows, or any task where accuracy matters.
GitHub Workflows & Version Control
Store your workflows and tools in GitHub so you have version history, can collaborate with others, and can easily deploy to production. Claude Code integrates with GitHub, so you can manage branches, pull requests, and deployments.
Mindset: Genuine Curiosity
Approach Claude Code with genuine curiosity. If you're confused, ask it. If you wonder if something's possible, ask. If you want to understand how to build something, ask follow-ups. Claude Code is your best friend for learning technical concepts because it knows more than you and can explain anything.
Real-World Success Story: Christian's Path
From Cold Outreach to Warm Conversations
Christian started with high-volume cold email blasts (500+ per day) with generic AI templates. This failed because he was just another commodity seller. He switched to warm outreach, genuine conversations, and positioning himself as a partner who diagnoses problems. This led to his first client in 5 days.
Building a PM Assistant Workflow
Christian's first client was a construction company struggling with client communication. He built a PM assistant that let project managers talk to a bot, which would send client updates, log CRM entries, and answer FAQs. Inspired by Nate's AI assistant video, he adapted it to the client's specific pain point and delivered real value.
Honesty & Transparency Build Trust
Christian told clients 'I'm new, I'm learning, but I want to help you.' This honesty made him sound like a real human, not a salesperson. Combined with delivering value, it built trust fast. He also positioned himself as a guide through their AI journey, not just a template seller.
Continuous Improvement Through Feedback
After delivering the PM assistant, Christian met weekly with the client to refine and improve it. Each iteration made it better. This ongoing relationship turned a one-time project into a deeper partnership and opened doors for more work.
Confidence Comes from Reps
Before the community, Christian had no idea how to talk to people, price work, or deliver solutions. After following the framework and landing his first client, his confidence skyrocketed. He went from 'should I charge?' to 'here's my clear offer and ROI.' Reps and real results build confidence.
Notable quotes
Agentic workflows are not just a trend. They're the future of the AI industry. — Nate Herk
You're not charging for hours. You're charging for outcomes. — Nate Herk
Act as the doctor, not the pharmacist. Find the clog, clear it, then add more water. — Nate Herk
Action items
- Download and install Visual Studio Code, then install the Claude Code extension.
- Sign up for a paid Claude subscription (Pro at minimum, Max recommended for serious use).
- Create a cloud.md file for your first project using the WAT framework template from the community.
- Build your first agentic workflow using the newsletter example or a simple automation relevant to your business.
- Map out your trust network: list 20 people you know who run businesses or are in decision-making roles.
- Have 5-10 warm discovery conversations with people on your trust map to identify their pain points.
- Propose a free pilot workflow to the person with the clearest, most painful repetitive task.
- Build and test the MVP workflow, then gather feedback and iterate.
- Document your workflow with clear naming, sticky notes, and a Loom walkthrough video.
- Create a written scope of work and agreement before handing over any workflow to a client.
- Set up a maintenance retainer offer (10-25% of project cost per month) to create recurring revenue.
- Track and measure the impact of your delivered workflows (time saved, errors reduced, revenue generated).
- Join the AI Automation Society community to connect with others building AI businesses.