Bolt.new

Overview

Bolt.new by StackBlitz is an AI-powered full-stack development environment that runs entirely in the browser. Describe what you want to build and it writes the entire app — frontend, backend, database integration — and runs it live in a browser-based container.

My Experience with Bolt.new

I've spent a few months running Bolt.new through everything from weekend side-project ideas to a real MVP for a non-technical founder I was advising. As a browser-based idea-to-working-app machine, it's genuinely remarkable. As a tool for building something you'll run in production, it isn't trying to be one — treating it that way is where people get burned. Here's the honest breakdown.

The whole app lives in your browser — no local setup, ever

The thing that still surprises people the first time is that there is nothing to install. Bolt.new runs your entire app — frontend, backend, package installs, dev server — inside a StackBlitz WebContainer, a sandboxed Node.js environment executing directly in the browser tab. A live preview sits right next to the chat, so the moment the AI finishes writing code, you're looking at a running app, not a pile of files to piece together. For testing an idea on a borrowed laptop, or anywhere you don't want to touch your local Node setup, that's genuinely useful in a way desktop-first tools aren't.

Be specific upfront — the token budget punishes vague prompts

This is the lesson that took me longest to learn. Bolt's free tier runs on a daily token allowance, and every regeneration, every "actually, also add X" follow-up, eats into it. Start vague and iterate feature-by-feature in chat, and you'll burn through the budget before anything is usable. What works instead: one long, detailed initial prompt spelling out the full feature set — auth method, exact pages, rough database schema, styling — all at once. I've gotten a working task app with user auth from a single detailed prompt in under 30 minutes that way. Specificity isn't about quality here, it's about not running out of tokens.

Export to local, then hand off to Cursor for the real refinement

Bolt has an export-to-local option that quietly makes it more useful than it first appears. Once a prototype is in decent shape, I export it and open it in Cursor for the parts that need real engineering judgment — tightening error handling, fixing duplicated logic, adding the tests Bolt never writes on its own. Bolt as the fast first draft, Cursor as the editing pass, is the right division of labor. Refining everything inside Bolt's own chat is slower and burns tokens better spent on new features.

Built for non-technical founders who need to ship before raising money

The use case where Bolt comes closest to transformative is a non-technical founder who needs a clickable, working product to show investors, not a slide deck. I've watched someone with zero coding background go from an idea to a deployed prototype with real signups and a database in an afternoon. That changes what "MVP" means for people who previously had to hire a developer just to validate an idea. It won't be architected the way a senior engineer would build it, but it's real, it runs, and it's enough for first feedback.

Where it's genuinely great: demos and proof-of-concept

Beyond MVPs, Bolt is the tool I reach for to prove a concept has legs before committing engineering time — internal tool ideas, "could we build X" conversations, client pitches. Because the backend is real and not mocked, a demo built in Bolt shows actual data flowing through actual database calls, which lands differently with stakeholders than a static click-through. I've spun up a working proof-of-concept mid-meeting while someone was still describing the requirement — hard to pull off with any other tool I've used.

Honest weaknesses

Every piece of code Bolt generates needs a careful review pass before it's anywhere near production — it writes working code, not necessarily secure or maintainable code, and input validation and error boundaries are often thin. Context confusion is real once a project grows: past a certain number of files and features, Bolt loses track of earlier decisions and will contradict its own previous architecture. You also get very little say in the underlying structure — it picks the stack and patterns, and disagreeing usually means asking it to redo work rather than steering it directly. If architectural control matters to you, that will frustrate you.

Bolt.new vs v0: full apps vs components

These get compared constantly, but they solve different problems. v0 focuses on polished UI components for React/Next.js projects you already have — it's excellent at "build me a pricing page" or "make this form look better." Bolt targets the whole app: frontend, backend, and database together, running live, from a blank slate. Already have a codebase and need a good-looking component dropped in? Use v0. Starting from nothing and need a full-stack prototype today? Bolt is built for that job.

Is Bolt Pro or Team worth the money?

The free tier's daily token allowance is fine for evaluating the tool or building one small prototype, but it runs out fast once you're iterating on a real project — you'll hit the wall mid-session more than once while learning to write specific prompts. Pro at $20/mo removes that anxiety and is worth it once Bolt is more than a one-off experiment for you. Team at $50/mo makes sense once more than one person needs shared usage rather than everyone burning their own token pool. For a solo founder validating one idea, Pro is the right tier — don't pay for Team until there's an actual team.

Pricing