Vibe Coding for Complete Beginners
You do not need to become a traditional programmer before you can build something useful. You do need a clear problem, a willingness to test what the AI produces, and enough basic knowledge to notice when the result is wrong. This series teaches exactly that.
Vibe coding means building software through an ongoing conversation with an AI coding agent. You describe the outcome, the agent studies the project, proposes a plan, edits files, runs commands, and helps you verify the result. It is more powerful than copying a snippet from a chatbot because an agent can work across a real project. It is not magic: the human still decides what should exist and whether the result is acceptable.
What you will build
You will complete three connected projects:
- A responsive web expense dashboard with Next.js, TypeScript, Tailwind CSS, Supabase, and Vercel.
- A mobile expense tracker with Expo and React Native that runs on Android and iOS.
- A standalone expense API with Python, FastAPI, PostgreSQL, authentication, tests, and monitoring.
The projects deliberately share one business domain. Once you understand a transaction, category, budget, and user in the web project, you can recognize the same ideas on mobile and in an API. You spend less energy inventing features and more energy understanding how the layers fit together.
What the AI does—and what you do
The AI is good at creating boilerplate, explaining unfamiliar files, connecting established libraries, writing tests, and investigating error messages. It can also make confident mistakes, use outdated APIs, expose secrets, or implement a different product from the one you intended.
Your job is to provide five things:
- Outcome: what the user should be able to do.
- Context: which project and files are relevant.
- Constraints: technologies, accessibility, security, and scope.
- Acceptance criteria: observable conditions that define “done.”
- Verification: commands and manual checks that prove the change works.
You are not expected to understand every line immediately. You are expected to ask for explanations, review the visible behavior, run the checks, and preserve a working version before a risky change.
Choose your AI coding tool
This course supports Codex, Claude Code, and Cursor. Pick one as your daily tool; do not pay for all three at the beginning.
- Codex is an agent for exploring codebases, planning changes, editing files, running checks, and reviewing work across desktop, command-line, editor, and cloud workflows.
- Claude Code is an agentic coding environment available in the terminal and other surfaces. It can read files, edit them, run commands, and use project instructions.
- Cursor is an AI-first editor based on familiar code-editor workflows. Its Agent can inspect a project, make multi-file changes, and run commands.
The prompts in this course are tool-neutral. A note labeled with a tool name only appears when its interface or setup differs.
The safe working loop
Use this loop for every feature:
- Describe: write the user outcome and boundaries.
- Explore: ask the agent to inspect the current project without editing.
- Plan: request a short plan, affected files, risks, and tests.
- Change: authorize one small, reviewable slice.
- Verify: run linting, type checks, tests, and a manual user flow.
- Review: inspect the changed files and ask what remains uncertain.
- Save: create a Git commit when the project works.
This loop is slower than “build the whole app” in one prompt and much faster than repairing an unreviewable pile of generated code.
Your first tool-neutral prompt
Create an empty folder named vibe-coding-lab, open it in your chosen tool, and send:
I am a complete beginner. Inspect this folder without changing anything.
Explain what you can do in this environment, what you cannot verify for me,
and how you will ask for approval before risky commands. Then propose one tiny
Markdown file exercise that teaches me how to review a change.Review the proposal. If it is safe, ask the agent to create hello.md with a heading, a two-sentence project goal, and a checklist. Open the file yourself. Change one word manually. You have now completed the essential loop: request, review, inspect, and edit.
Costs and expectations
You can learn many parts for free: VS Code, Git, Node.js, Python, Next.js, Expo, FastAPI, PostgreSQL, and local development are free. AI coding products and hosted services have usage limits or paid plans. Publishing to app stores can require developer-account fees. Prices change, so Article 3 links to the current official pricing pages instead of treating a price as permanent.
Do not buy infrastructure before the project needs it. The web and database projects can begin on free tiers, and the mobile app can run on your phone during development. Set spending limits wherever a provider offers them.
How to know you are progressing
Progress is not measured by generated lines of code. At the end of each lesson, you should be able to answer:
- What user behavior did I add?
- Which files changed, and why?
- How did I verify it?
- What data or secret could be at risk?
- Can I return to the last working version?
If you cannot answer, stop and ask the agent to explain the change in plain English. Confusion is a signal to reduce the size of the next step.
Common beginner traps
Starting with a giant marketplace or social network. Begin with one user and a few well-defined records. Complexity multiplies when you add roles, payments, chat, maps, or real-time collaboration.
Changing stacks after every video. Choose one path and finish it. Alternatives make more sense after you have shipped once.
Pasting secrets into chat or source code. Use environment variables, commit an example file without real values, and rotate any credential that was exposed.
Accepting “it should work.” Ask the agent to run the relevant checks and tell you what it could not test.
Skipping Git. A working commit is the safest undo button you have.
Completion checklist
- I can explain vibe coding without calling it automatic app generation.
- I selected one primary AI coding tool.
- I completed the
hello.mdexercise. - I understand the seven-step working loop.
- I will not paste real credentials into prompts or committed files.
Next: What You Can Build With AI Coding Tools