Bolt.new Review: What Actually Works and What Doesn't

Yes, Bolt.new is legit — but with important caveats. We spent a week pushing StackBlitz's AI-powered dev sandbox to its limits, and it genuinely builds and runs full-stack apps in-browser from a prompt. That said, it has clear boundaries that every developer should know before subscribing.

Why Bolt.new Actually Delivers

The promise of AI web development is usually a lie. Most "prompt-to-app" tools dump broken React code on you and expect you to figure out the rest. You still have to configure Webpack, install Node modules, and debug obscure terminal errors.

Bolt.new solves that core problem. It spins up an entire development environment directly in your browser. It installs packages, configures databases, and deploys your application with a single click.

What Makes Bolt.new Different?

To understand Bolt.new, you have to understand StackBlitz's core technology: WebContainers. Unlike other AI tools that run on a remote server, Bolt.new runs a Node.js environment directly inside your browser tab. When you ask Bolt to build a full-stack Next.js app, it doesn't just output a code block — it actually runs the local server in your browser.

Key features that stood out:

In-Browser Terminal Access: See actual build steps, package installations, and console logs. If something breaks, you can manually run npm commands in the terminal right next to the chat.
Full-Stack Capability: Not restricted to static HTML. It runs SQLite databases, backend APIs, and modern frameworks like SvelteKit, Vite, or Next.js.
Auto-Correction Engine: If the build fails, Bolt.new reads the terminal error and automatically patches its own mistake. That feedback loop is genuinely satisfying.
GitHub Integration: One click pushes the generated codebase to a fresh repository, ready for local development.

Real-World Test: Building a Kanban Board

We didn't ask Bolt to make a basic to-do list. We pushed it to build a functioning, multi-user Kanban board with drag-and-drop mechanics, persistent storage, and an export-to-CSV feature.

The setup took exactly two prompts:

> "Build a Trello clone using Tailwind CSS and Lucide icons. Store data in localStorage, support drag-and-drop, and add a quick export button."

Within 90 seconds, the app was fully functional. The drag-and-drop library installed, the UI was clean, and the preview was interactive. When we asked for card-description editing, it updated existing files without breaking the state management.

For rapid prototyping and building MVPs, this workflow is hard to beat. You can test an idea in ten minutes instead of spending half a day configuring boilerplate.

Where Bolt.new Falls Flat

No tool is magic. Bolt.new has serious limitations that keep human engineers relevant.

Context window drift is real. As your project grows, Bolt begins to forget earlier design choices. On a project with 15+ files, it will occasionally overwrite critical utility functions and break your app during major architectural changes.

Token burn is expensive. Every prompt forces the underlying LLM (usually Claude 3.5 Sonnet) to read your entire file tree. Large projects consume subscription credits quickly.

Complex external integrations are tricky. Setting up OAuth with Google or integrating a Stripe webhook is difficult because Bolt can't test live external endpoints inside its sandboxed browser environment. You'll still need to pull code locally for the heavy lifting.

Should You Buy It?

Bolt.new is not a replacement for a seasoned software engineer. If you try to build a complex, enterprise-grade system with hundreds of routes, you will hit a wall.

However, it is easily the best rapid prototyping environment on the market today. If you are a solo founder launching an MVP, an engineer learning a new framework, or a designer building functional mockups, Bolt.new is worth the subscription. It cuts out the boring setup phase and lets you focus entirely on the product.