# Vibe coding for beginners: from zero to your first app

> Vibe coding for beginners comes down to one loop: describe what you want to an AI coding agent in plain words, let it build in a separate copy of your project, look at the result, and keep only what you reviewed. You need a computer, Git, one agent account and a small first idea. This guide takes you from zero to a first working app.

- Canonical: https://runtallos.com/learn/vibe-coding-from-scratch
- Português: https://runtallos.com/pt/aprenda/vibe-coding-do-zero.md
- Author: [Roberto Rocha](https://www.instagram.com/robertorochamkt/)
- Updated: 2026-09-30
- Section: Learn

## Key facts

- No professional coding background needed to start
- One agent account, subscription or API key, is enough
- Agents ask before running commands by default
- Every change lands in a separate copy you can discard
- Start from an empty folder: Tallos creates the Git project

## What is vibe coding, and can a beginner do it?

**Vibe coding** is building software by describing what you want to an AI coding agent in plain language and letting it write the code. Yes, beginners can do it. The agent reads files, writes code and runs commands, so your job becomes explaining the goal, checking the result and deciding what to keep.

What beginners lack is rarely typing speed; it is judgment about what the agent changed. This guide builds that habit from day one. For the concept and where it breaks down, read [what vibe coding is](https://runtallos.com/vibe-coding).

## What do you need to start vibe coding?

You need four things, and none of them is a computer science degree.

- **A computer** running macOS 13 or later (Apple Silicon or Intel) or Windows 10 or later.
- **Git.** It records every version of your project and powers the separate copies agents work in. Tallos needs it to create projects and workspaces.
- **One AI coding agent account**: a subscription or an API key with a vendor such as Anthropic (Claude Code), OpenAI (Codex) or Google (Gemini CLI).
- **A small first idea.** A to-do list, a page that tracks your reading, an expense tracker. Something you can check in a minute.

Tallos is the workspace around the agent, an [agentic development environment](https://runtallos.com/what-is-an-ade). It installs and signs in agents, gives every task its own copy of your project and shows you every change before you keep it.

## Which AI coding agent should a beginner pick?

Pick the agent whose account you already have or can set up today; you can add others later. Coding agents are command-line programs (CLIs) from different vendors; Tallos runs all of these.

| Agent | Made by | How you sign in |
| --- | --- | --- |
| Claude Code | Anthropic | Account login or an Anthropic API key |
| Codex | OpenAI | Account login or an OpenAI API key |
| Gemini CLI | Google (open source) | Google sign-in or a Gemini API key |
| GitHub Copilot CLI | GitHub | Your GitHub account, as GitHub describes |
| OpenCode | Open source, many model providers | The provider logins you set up in OpenCode |
| Grok CLI | xAI | Grok login or an xAI API key |
| Qwen Code | Alibaba's Qwen team (open source) | `/auth` inside Qwen Code |
| MiMo Code | Xiaomi | `mimo auth login` |

_Pricing and free usage change often; check each vendor's page before you choose._

All but Copilot CLI have a card in Settings → Connections that installs the agent and opens its sign-in. See [how to connect your agents](https://runtallos.com/learn/connect-your-agents).

## How do you vibe code your first app, step by step?

Do these nine steps once on a small idea; every later feature follows the same loop.

1. **Install Tallos** — Download Tallos for macOS or Windows and open it. On the welcome screen, pick your language and your agent under **Which AI agent do you use?** Leave **Let agents run commands without asking** off: the agent will ask before it runs anything.
2. **Connect your agent** — Open Settings → Connections. On the agent's card, install it, then sign in or save an API key. A key you save there goes only to that agent.
3. **Create a project** — Starting from nothing? Open **Add a project** and choose **Create new project**. Name it, and Tallos creates a folder that is already a Git repository. Have code already? **Choose a folder** or **Clone from a URL**.
4. **Start a task in its own workspace** — Click **New workspace**, pick the project and the agent, and leave the name blank if you like. Tallos works in a separate copy of your project, so your original stays safe.
5. **Describe the first version** — Write what the app is, who uses it and the one thing it must do. Ask for the simplest version that runs, and ask the agent to tell you how to start it.
6. **Answer the agent** — The agent may ask a question or ask permission to run a command, such as installing a package. Read what it wants to run. Allow it if it matches your task; otherwise say no and explain.
7. **Look at the result** — Start the app the way the agent explains and open its address in the workspace's built-in browser. Click around and note what is wrong or missing.
8. **Iterate in small steps** — Ask for one change at a time: “make the Add button bigger”, “save the list when I close the page”. When words fail, click an element in the browser and send it to the agent.
9. **Review before you keep** — Click **Review changes** to see every file the agent touched. Leave notes on lines you don't understand and send them back. When it looks right, commit it. If the attempt went wrong, delete the workspace and start again.

[Your first task in Tallos](https://runtallos.com/learn/first-task) walks the same loop in more detail, with troubleshooting.

> Try the loop on your own idea. Tallos runs on macOS and Windows with the agent you already use. → https://runtallos.com/signup

## How do you describe what you want to an AI agent?

Describe the outcome, the context and how you will check it. Agents do well with concrete requests and poorly with vague ambitions.

**Vague prompt**
- “Build me a budgeting app.”
- “Make it look better.”
- “Fix the bug.”

**Clear prompt**
- “A one-page app where I type an expense and a category and see a monthly total. Keep the data in the browser, no login.”
- “Make the total bigger and bold, and put the form above the list.”
- “When I type 3,50 as an amount, the total shows NaN. Fix it and add a test.”

- Say **who** it is for and **the one thing** it must do first.
- Name what to leave alone: “don't change the layout”.
- Ask for the plan before a big change.
- End with how you will check it: “the total should update when I add an item”.

## How do you iterate without breaking what works?

Iterate in small, separate steps and save each working version. The fastest way to lose a working app is to ask for five things at once.

- **One change per message.** Small changes are easy to check and easy to undo.
- **Commit what works.** A commit is a save point. Keep iterating in the same workspace and commit each working version; its branch holds your app until you merge it, for example through a pull request.
- **Ask for tests.** Tell the agent to add a test for what it changed, so it can check its own work.
- **Start fresh when it spirals.** If the agent goes in circles, stop arguing. Delete the workspace and start a new one with a clearer prompt.

## How do you review AI-generated code if you can't read code well?

You review by checking scope, behavior and anything you don't understand. You don't need to read every line fluently to catch most problems.

- **Scope:** which files changed? A “small fix” that touches many files is a warning sign.
- **Behavior:** does the app do what you asked, and do the tests pass?
- **Questions:** leave a note on any line you don't understand, such as “what does this do?”, and send your notes to the agent. Its answers teach you the code.
- **Secrets:** no passwords or API keys written into the code.

When you are ready for a stricter pass, use the checklist in [how to review AI-generated code](https://runtallos.com/learn/review-ai-generated-code).

## What are the most common beginner mistakes?

| Mistake | What happens | Do this instead |
| --- | --- | --- |
| Asking for the whole app at once | A huge change you can't check | Build the smallest version, then add one feature at a time |
| Keeping changes without looking | Bugs pile up unnoticed | Open **Review changes** every time |
| Turning off permission prompts on day one | An agent can edit or delete files without checking with you | Stay in manual mode until you trust the task |
| Pasting passwords or keys into prompts | Secrets end up in code or history | Keep them in environment files that Git ignores |
| Arguing with a stuck agent | Long sessions that go in circles | Discard the workspace and restart with a clearer prompt |
| Never committing | No save point to go back to | Commit each working version |

_Beginner mistakes and how to avoid them_

## How do you keep vibe coding safe?

Three defaults keep beginners safe in Tallos: agents ask before acting, every task runs in a separate copy, and Git remembers every version.

- **Manual permissions by default.** Tallos starts agents in manual mode, so they ask before running commands. Autonomous mode is opt-in, under Advanced on the welcome screen or in Settings → Agents → Agent Permissions.
- **Isolated workspaces.** Each task is a git worktree with its own branch. The agent edits that copy, never your original, and deleting the workspace removes the folder and its branch.
- **Git history.** Every commit is a save point. You can always get back to the last version that worked.
- **Keys stay with their agent.** An API key you add in Settings → Connections reaches only that agent.

> **A workspace is not a sandbox** — A workspace keeps your tasks apart; it does not stop an agent from touching the rest of your computer. That is what permission prompts are for: read them before you allow.

## What comes after your first app?

Once the loop feels natural, run more than one agent. Parallel agents and squads are the same loop, multiplied.

- **Two agents, one task.** Click **Try another agent** to give the same task to a second agent in its own copy, then keep the better result.
- **Different tasks at once.** One agent on the page layout, another on saving data, each in its own workspace. See [how to run AI agents in parallel](https://runtallos.com/run-ai-agents-in-parallel).
- **A squad.** A leader agent splits an objective across member agents, checks their work, asks you only when blocked and stops so you can accept or discard. Start with the **Build + review** template: one agent builds, another reviews. Follow [your first squad](https://runtallos.com/learn/your-first-squad).

> Get Tallos and start with one agent and one small idea. → https://runtallos.com/signup

## Frequently asked questions

### Can I vibe code with no programming experience?

Yes. You can build small apps by describing them to an AI coding agent. You still need to check the result: run the app, look at which files changed and ask the agent to explain anything unclear before you keep it.

### How do I vibe code an app from scratch?

Create a new project, start a task in its own workspace, describe the smallest useful version of the app, run it and ask for one change at a time. Review and commit each working version.

### Which AI agent is best for vibe coding beginners?

There is no single best one. Start with the agent whose account you already have, such as Claude Code, Codex or Gemini CLI. Tallos runs all of them, so you can compare them later on the same task.

### Do I need a new AI subscription to use Tallos?

No. Tallos uses the agent subscriptions or API keys you already have and does not resell model access. Check each agent vendor's page for its own pricing.

### Is vibe coding safe?

Yes, with three habits: agents ask before running commands, you work in a separate copy and you review changes before keeping them. Tallos makes all three the default.

### What should my first vibe coding project be?

Something small you can check in a minute: a to-do list, an expense tracker, a page that lists your books. Build the smallest version first, then add one feature per task.

## Sources

1. [“There's a new kind of coding I call vibe coding” (Feb 2, 2025)](https://x.com/karpathy/status/1886192184808149383) — Andrej Karpathy on X
2. [Best practices for Claude Code](https://code.claude.com/docs/en/best-practices) — Anthropic
3. [Claude Code overview](https://code.claude.com/docs/en/overview) — Anthropic

---

**Run your agents in parallel with Tallos** — Claude Code, Codex, Gemini and 30+ agents — each in its own workspace, with squads, chat, terminals and review in one app. Uses the subscriptions you already have.

https://runtallos.com/signup (macOS 13+ · Windows 10+)