How to Vibe Code: Getting Started Using AI Coding Tools

In February 2025, AI researcher Andrej Karpathy posted a description of a new way he was building software: leaning fully into AI tools, barely reading the code it generated, just vibing with the output until it worked. He called it "vibe coding." The term stuck because it captured something real: a shift where describing intent clearly matters more than writing syntax correctly.

This isn't a trend that's going to fade. Every major AI coding tool (Cursor, Bolt, GitHub Copilot, Replit) has gotten dramatically more capable in the year since. Developers use them to move faster. Non-developers use them to build things that would have required hiring an engineer eighteen months ago.

If you've been watching people ship apps on social media and wondering how they're doing it, this guide breaks it down: the tools, the process, and what to build first.

What Is Vibe Coding?

Vibe coding is building software the way you'd describe a task to a really capable colleague: in plain language, iteratively, without worrying about the implementation details. You say what you want. The AI writes the code. You review the output, push back on what's wrong, and refine until it works.

The key shift is what you're responsible for. You're not responsible for knowing the right function name or syntax. You're responsible for having a clear idea of what you're building, being able to spot when the output doesn't match your vision, and knowing how to describe the gap back to the AI so it can fix it.

Vibe coding isn't about ignoring code. It's about delegating the writing of it, while staying in the driver's seat on what it does.

This distinction matters. The people who get stuck in vibe coding sessions are usually the ones who accepted a broken output without pushing back, or who asked for something vague and got something vague back. Clear input, sharp feedback: that's the whole skill set.

The Tools That Work

There are a lot of AI coding tools right now. Here's how they break down by what you're trying to do, so you're not wasting time figuring out the ecosystem before you've built anything.

Start with zero setup

These tools run in the browser. You don't install anything, configure an environment, or touch a terminal. For first-time builders, this is where to start.

• Bolt: describe an app in plain English and it generates a full working prototype. Strong on React and web apps. One of the fastest from idea to running thing.
• Replit: a browser-based coding environment with AI built in. Good if you want to see the code, run it, and iterate on it in one place. Has a real deployment button.
• v0 by Vercel: generates UI components from text descriptions. Best for people building front-end interfaces who want something that looks good immediately.

In-editor AI tools (more control, more power)

Once you're comfortable with the loop of describe → review → refine, these tools give you more control over what you're building and how.

• Cursor: the most popular AI-native code editor right now. It understands your whole codebase, not just the file you're looking at. Most developers who've switched to it don't go back.
• GitHub Copilot: Microsoft's AI pair programmer. Lives inside VS Code. According to GitHub's own published research, developers complete tasks 55% faster with Copilot active. If you already have a codebase, this integrates without disrupting your setup.
• Windsurf: newer entrant, strong on multi-file edits and understanding context across a project. Worth knowing if you outgrow Cursor's free tier.

Your First Vibe Coding Session

The first session is where most people either get hooked or give up. The difference is almost always how they framed the initial prompt, not the tool they chose or their technical background.

Here's how a good first session runs:

1. Write your spec before you open the tool. One paragraph. What is the app? What does it do? Who uses it? What does "done" look like? This sounds simple but most people skip it and spend an hour refining a vague output.
2. Give the AI your full spec as the first prompt. Don't ease in: dump everything in the first message. AI coding tools perform better with full context upfront.
3. Review the output against your spec, not your instincts. Ask: does this do what I said I wanted? Not: does this look like what I imagined? The spec is your source of truth.
4. Give specific feedback, not general feedback. "The button doesn't work" is hard to fix. "When I click Submit, nothing happens, it should save the entry and show a confirmation message" is fixable in one generation.
5. Ship something small before you add features. The instinct to keep adding is where first builds die. Get one working thing deployed before you expand scope.

What to Build First

The best first project is one you need. Not a tutorial clone, not a toy app: something that solves a problem you run into regularly. Personal projects ship faster because you know what "done" looks like without having to guess.

Here are project types that work well for first-time vibe coders:

Avoid anything with payments, complex user auth systems, or real-time data on your first build. These are solvable: they just add surface area that makes it harder to know whether a bug is in your logic or in the infrastructure you haven't fully set up yet.

Where BYOBot Fits In

The single biggest accelerant in a vibe coding session is walking in with a clear spec. And writing a good spec: especially for an automation, a workflow tool, or anything that needs to talk to other apps: is harder than it looks.

BYOBot is built for exactly this step. Describe what you want to automate or build, and BYOBot produces a structured playbook: what the system does, what triggers it, what it outputs, and how the pieces connect. You take that spec straight into Cursor or Bolt and the AI coding tool has everything it needs to generate something accurate on the first try.

This is especially useful if you're building something that touches existing tools: a workflow that reads from a spreadsheet, posts to Slack, updates a CRM. The spec has to be tight, or the AI will make assumptions you didn't intend. BYOBot handles the spec so you can focus on building. You can see how the AI workflow design process works here, or just open it and describe what you're making.

A vague prompt produces a vague app. The five minutes you spend writing a proper spec saves three hours of "why isn't this working."

Frequently Asked Questions

• What is vibe coding? +
  Vibe coding is the practice of building real software (apps, tools, scripts, automations) by describing what you want to an AI coding tool and iterating on what it produces, rather than writing code line by line yourself. The term was coined by AI researcher Andrej Karpathy in early 2025 to describe a shift where intent replaces syntax as the core skill. You describe; the AI writes; you review and refine. The loop is the whole job.
• Do I need to know how to code to vibe code? +
  No. That's the whole point. You need to be able to describe what you want clearly, spot when the output doesn't match your intent, and know how to push back with specific feedback. Those are communication skills, not coding skills. Some basic familiarity with how apps are structured helps, but you can pick that up as you go. Start with a zero-setup tool like Bolt or Replit, build something small, and you'll have more context in an hour than any tutorial gives you in a week.
• Which AI coding tool should I start with? +
  For building full apps from scratch with no setup, start with Bolt or Replit: both run in the browser and get you to a working prototype fast. For working inside a code editor with more control, Cursor is the tool most developers use daily. GitHub Copilot is the right call if you already have a codebase you're working in. If you want help figuring out which tool fits your specific build, BYOBot can walk you through it based on what you're making.
• What should I build first? +
  Build something you need: a tool that solves a real problem in your life or work. Personal productivity tools, portfolio sites, simple dashboards, and workflow automations are all strong first projects. Avoid anything that requires user authentication, payments, or complex databases on your first build. Those are solvable: they just add complexity before you've got the core loop down. Ship something that works, then expand.