$ man content-wiki/repos-and-skills-for-builders
Content Workflowsbeginner
the repos and skills you actually need to build AI-powered apps
You don't need to be a senior developer. you need repos, basic CLI, and how to work with AI coding assistants.
by Shawn Tenam
the stack reality
The barrier to building production AI apps dropped significantly in 2024-2025. AI coding assistants can write code you don't fully understand yet, which means you can build things before you understand all of them. That's a feature, not a cheat.
What you actually need: the ability to run terminal commands without panicking, basic understanding of what a repository is and why it exists, and enough Next.js familiarity to understand what you're reading when Claude explains it. The rest you learn while building.
What you don't need to learn first: Docker, Kubernetes, AWS infrastructure, advanced TypeScript generics, database optimization, or system architecture theory. Learn those when a specific problem requires them. Build something first.
git fundamentals for builders
Git is version control. It tracks every change to your code so you can undo mistakes and collaborate without overwriting each other's work. The five commands that cover 90% of daily use:
git status ... see what files have changed
git add ... stage a file for commit
git commit -m "what you did" ... save a checkpoint
git push ... send your changes to GitHub
git pull ... get the latest changes from GitHub
The other 10%: git log (see history), git branch (create a parallel version), git checkout (switch between branches), git merge (combine branches). You'll learn these when you need them.
One rule: never commit .env files. They contain API keys and credentials. Add .env to your .gitignore file and it will be excluded automatically. If you accidentally commit one, rotate the key immediately ... it's now public.
monorepo architecture
A monorepo is one repository that contains multiple projects. The shawnos.ai setup runs three separate websites from one codebase. Shared components (navigation, buttons, utility functions) live once and all three sites use them. Update the component once and all sites get the update on next deploy.
Turborepo and Nx are the two main monorepo management tools. Turborepo is simpler for most builders and has excellent Vercel integration (they're the same company). Nx has more advanced features that become relevant at team scale.
The practical benefit: if you're building more than one site or app, start with a monorepo structure even if it feels like overkill. Migrating into a monorepo later is painful. Starting in one and having everything connected from day one is a real time saver.
Next.js, Vercel, and the deploy pipeline
Next.js is the production framework for this stack. Server components handle data fetching without exposing API keys to the browser. API routes let you build backend logic without a separate server. The app router is the current standard ... use it for any new project.
The docs at nextjs.org are genuinely good. The "Learn Next.js" tutorial at nextjs.org/learn takes about 4-6 hours and covers 80% of what you need for production apps. Do the tutorial before trying to build anything complex.
Vercel is the deployment layer. Connect your GitHub repo and every push to main automatically builds and deploys your site. No configuring CI/CD pipelines, no managing servers. The free tier handles most personal projects. The killer feature for Next.js is edge functions ... serverless functions that run close to users globally with no configuration.
Deploy cadence matters for learning. The tightest feedback loop is: write code locally, push to GitHub, watch Vercel build logs, see the result live. That loop should take under 2 minutes. If it's taking longer, something in the pipeline needs fixing.
Claude Code CLI and Cursor (the AI layer)
Claude Code CLI is an AI agent that lives in your terminal and operates on your entire codebase. Unlike a chat interface where you paste code snippets, Claude Code reads all your files, understands the architecture, and can write code that fits your existing patterns. It can run tests, commit changes, and manage the repo.
The nightly cron pipeline uses Claude Code to run automated content operations while you sleep. A cron job triggers Claude Code, it reads the repo state, executes the task (generating content, updating data files, running reports), commits the results, and pushes. No human in the loop.
Cursor IDE is the interactive complement to Claude Code. When you want to write code alongside AI explanation and iteration, Cursor is the environment. The codebase-aware chat can explain any file, suggest refactors, and debug errors in context. The inline completions are faster than switching to a chat window.
MCP (Model Context Protocol) is the protocol that lets AI agents connect to external tools: browsers, databases, APIs, task managers. MCP servers expose capabilities that Claude Code and Cursor can call. The multi-agent system runs on MCP connections between tools ... Claude Code coordinates with Playwright for browser automation, with Notion for knowledge base updates, with Attio for CRM operations.
the learning path
The sequence that minimizes wasted time:
1. Git basics: commit, push, pull, .gitignore. One afternoon. Use GitHub's own tutorial or the git-scm.com book's first three chapters.
2. Next.js tutorial: nextjs.org/learn. Do the whole thing. 4-6 hours. Build the demo app.
3. Deploy something: take your tutorial app, connect it to Vercel, push to GitHub, watch it deploy live. This step matters for confidence.
4. Add Claude Code: install the CLI, run it on your repo, ask it to add a feature. Watch how it reads the codebase and writes code that fits.
5. Build a real feature: something you actually want. The tutorial app won't teach you what building for yourself teaches you.
The learning trap to avoid: spending weeks on documentation before writing any code. The documentation is for when you're stuck, not for prereading. Build, break, get stuck, look it up. That sequence is faster than the alternative.
frequently asked questions
Q: Do I need to know JavaScript before TypeScript?
A: TypeScript is JavaScript with types added. Most tutorials start with JavaScript fundamentals and layer TypeScript on top. If you're starting from zero, the JavaScript.info tutorial is the best free resource. Give it 2-3 weeks of daily practice before jumping into Next.js.
Q: What about backend and databases?
A: Next.js API routes handle most backend needs for early-stage apps. For data, start with SQLite locally (zero config) and Supabase (Postgres with a free tier) when you need a hosted database. Skip self-hosted Postgres and ORMs until you have a specific problem that requires them.
Q: How much does this stack cost to run?
A: GitHub (free for public repos), Vercel (free tier covers most personal projects), Claude Code (subscription), Cursor (free tier is functional, $20/month Pro). Total for a real production stack: $20-40/month.
Q: Can I build a production app without any coding knowledge?
A: Claude Code and Cursor can write most of the code but you need enough literacy to understand what's happening, debug when things break, and make architectural decisions. The minimum viable coding knowledge is: read TypeScript and understand it, run terminal commands, read error messages and know where to look for help.
related entries