Bolt.new went from near-shutdown to $40M ARR in five months. I sat down with COO Alexander Berger to get the real story, and his take on how anyone can build software in 2025.
StackBlitz had 15 employees, less than $1M in annual recurring revenue, and was running out of reasons to keep going.
For four years, the company had been selling an in-browser code editor to a sliver of the developer market: design system teams at enterprises. The product worked. The technology was real. But the commercial traction wasn’t there. The team was actively discussing returning remaining capital to investors and shutting down.
Then, in October 2024, they launched Bolt.new with a single tweet.
No ad spend.
No PR campaign.
No launch event.
Just a tweet thread that COO Alexander Berger and CEO Eric Simons spent 12 hours crafting.
Five months later, Bolt.new hit $40M ARR [1]. It had become one of the fastest-growing software products in history, second only to ChatGPT [5].
I sat down with Alexander for a long-form conversation to understand what actually happened, how the product works for people with zero coding experience, and where he thinks this whole space is headed. Here’s what stood out.
What is Bolt.new and how is it different from Cursor and Lovable?
The AI coding tool space is getting crowded. Cursor, Lovable, Replit, v0, Windsurf. If you’re trying to figure out which one to use, Alexander’s framing is the clearest I’ve heard:
“We are easier to use than Cursor, especially for non-developer users. And we’re more powerful and enterprise-ready than the other vibe coding tools in the space.”
That positioning maps neatly onto how the market actually breaks down. Cursor is a VS Code-based IDE. It’s powerful, but it requires local installation, familiarity with developer environments, and comfort working in code. It’s built primarily for developers working in real codebases, and not typically recommended as a starting point for someone just discovering AI-powered development [3]
Bolt.new runs entirely in the browser. You go to bolt.new, type what you want, and hit enter. No IDE, no Git, no terminal, no local dev server. (If you don’t know what I’m talking about here, you should use Bolt!)
During our interview, Alexander typed a prompt for a calculator app with animated race cars and confetti effects. It was built and running in under two minutes.
Press enter or click to view image in full size
Lovable occupies similar territory to Bolt for non-technical users, but Alexander draws the distinction on the enterprise side. Bolt’s core technology, WebContainers, runs the entire development environment inside your browser tab rather than on a remote server [4]. For enterprise customers, this means proprietary code stays local rather than being processed on third-party infrastructure, and the platform scales to millions of users without spinning up millions of servers.
The other distinction, and this is where Alexander gets most animated, is enterprise code integration.
“In order for any of this AI generated code to get implemented at the enterprise, it needs to align with your enterprise code requirements, correct tech stack, security. And visually, it needs to be built using your company’s design system,”
Alexander explained.
“That is where a lot of these tools are really falling down and where we are here to help.”
Can someone with no coding experience actually build a real app?
I asked Alexander directly:
who is qualified to compose software with these tools?
“At this point, it’s extremely clear to me that anybody can do it. And if you don’t believe me, just DM me on Twitter and we’ll work it out together.”
But he was careful to distinguish between getting started and building something production-grade. An elementary school kid can make a game in Bolt right now. Alexander’s CEO’s mom made her first website with it. His wife built a site for her small business.
For more complex applications, Alexander says the skill isn’t coding. It’s what he calls “software composing”: the ability to describe what you want in the language of digital products.
“You want to get extremely good at describing very specifically what it is you’re trying to build in the language of digital products,”
he said.
“So if you don’t know the language of digital products, the beauty is AI will teach you everything that you need to know.”
He gave a practical example. If you want a navigation menu on mobile, you need to know it’s called a “hamburger menu” (or as Alexander calls it, a “pancake menu”). If you want space between elements, you should know the words “padding” and “margin.” These aren’t programming concepts. They’re design vocabulary. And knowing the right terms dramatically improves what the AI produces.
“It’s like in fantasy novels where with magic, you’ve got to know the name of the thing and that gives you the power over it. That is actually how it works in this space.”
For anyone looking to go beyond the visual interface into backend functionality, he recommends understanding what an API is (“basically how two pieces of software talk to each other”) and being able to describe data flows in plain language. Not how to build them. Just how to describe them.
Is Bolt.new good for beginners?
Yes, with a caveat. The tool itself has a near-zero learning curve: type and hit enter.
But getting consistently good results requires developing the vocabulary Alexander described above.
He also pointed to two features that most beginners don’t know about, which significantly improve the experience:
Visual inspector. A cursor-based tool that lets you select and modify specific elements of the UI directly. Alexander estimates only about 5% of Bolt users know this exists.
Plan mode. Instead of sending a prompt and hoping the AI interprets it correctly, plan mode forces the AI to generate a plan first. You review and revise the plan before approving it to build. This prevents the frustrating experience of the AI changing things you didn’t want changed.
Both features mimic how professional product teams actually work: inspect specific components, plan before building, and iterate in focused increments rather than broad strokes.
How did Bolt.new go from near-shutdown to $40M ARR?
The growth story is wild, but Alexander is clear that the “overnight success” was seven years in the making.
StackBlitz had spent those years building WebContainers, the technology that lets Bolt run a full Node.js development environment inside a browser tab. The technology worked, but it was trapped in a niche product (an in-browser code editor for developers). The commercial ceiling was low because most people can’t write code [4, 5].
Two things changed in 2024. First, Anthropic released Claude 3.5 Sonnet, which was the first AI model reliable enough to write code that would actually run without constant bugs. Second, the StackBlitz team realized they could hook that model into their browser-based environment and let anyone, not just developers, build software.
“We were lucky enough to stay alive long enough until the core technology that we built became commercially relevant,”
Alexander said. He credits the team’s discipline with cash. Under $1M in revenue for four years, but they ran lean, preserved capital, and didn’t blow money on things that didn’t directly improve the product.
The launch itself was a masterclass in focused execution. Alexander and Eric spent at least 12 hours constructing a single tweet thread. Most of that time went into the first tweet.
“We will agonize over those first 240 characters and do many iterations, trying to get just the most clear and succinct version of the simple idea.”
Alexander breaks down the exact formula in our full conversation: including why you should never put a URL in your first tweet, how to structure the rest of the thread, and what surprised them about who actually signed up. [Watch the full breakdown here.]
Press enter or click to view image in full size
Highlight: https://youtu.be/gcZXHRgSfi0?si=6d8HouZxu135CTZV
Full conversation: https://www.youtube.com/watch?v=8C3C4Vhknx8&t
Can you make money building apps with AI tools?
Alexander didn’t hedge on this one. His answer: yes, and here’s exactly how.
“This is where I’d put my money if I wasn’t working at Bolt and I needed to start a company with Bolt instead: agencies.”
The logic is straightforward. There are millions of small and medium businesses in America that could use much better software. Local services, restaurants, trades, small retailers. They’re not going to hire development teams. They need cheap, fast solutions.
What’s changed is how fast you can deliver. Alexander described the sales motion:
“In five minutes you can have a prototype of whatever it is you want to sell them with their marketing copy, their images off their website, the same sort of color pattern. You can really show instead of tell in the sales calls. So it’s never been easier to land a client.”
This aligns with a broader trend. AI automation agencies are emerging as one of the fastest-growing small business models, with some reaching $5,000 to $50,000 per month in revenue with minimal startup costs [6]. The difference with Bolt specifically is that you’re not just automating workflows. You’re building actual software products for clients: booking systems, lead capture tools, customer portals, internal dashboards.
What is vibe coding, and why does Alexander Berger dislike the term?
“Vibe coding” was coined by Andrej Karpathy in early 2025 to describe the practice of building software through natural language prompts rather than writing code directly [7]. The term went viral.
Alexander’s take:
it undersells what’s actually happening.
“I don’t like the term. It’s gotten a bad rap for the idea that you just throw half-baked prompts over the wall and let the AI figure it out.”
His preferred term is “software composing,” originally proposed by Riley Brown. The metaphor is deliberate: you’re orchestrating and composing an application from various inputs, ensuring the AI has the right context at every step. That’s a skill, not a vibe.
“It’s your job to ensure that the tool you’re using has the correct context loaded in at any given time to accomplish the task. And I think software composing captures that better.”
Whether the industry adopts his preferred terminology or not, the distinction matters practically. People who approach these tools as “vibing” get mediocre results. People who approach them as composing, with intentional context, vocabulary, and iteration, build real products.
How are enterprises using Bolt.new?
This was the most surprising part of the conversation for me. Bolt isn’t just for indie hackers and side projects. Alexander described an enterprise motion where Bolt literally deploys engineers into customer companies.
“We come in, we help integrate your enterprise code requirements so that your engineering team, instead of being maybe a bit apprehensive about this AI generated code, they’re thankful because it’s compliant and it uses the design system and it can actually be used and merged into the enterprise code base.”
The key insight: for AI-generated code to be useful at enterprise scale, it needs to match the company’s existing design system, tech stack, and security requirements.
Most AI coding tools produce generic output that engineering teams reject. Bolt’s approach is to customize the output to each enterprise’s specific standards.
Alexander acknowledged this isn’t a fully productized solution yet.
“I highly suspect that by next year we’ll have a very automated way of handling that. But yeah, at this point it’s us working hand in hand with our customers.”
The full conversation
This post covers the highlights, but the full interview goes much deeper into Alexander’s career path from first enterprise sale to COO, the specific frameworks for writing viral product launch tweets, his book recommendations for different life stages, and his perspective on how AI is reshaping team structures.
Frequently asked questions about Bolt.new and AI coding tools
What is Bolt.new? Bolt.new is an AI-powered development platform that lets you build, run, and deploy full-stack web applications through natural language prompts, entirely in the browser. It’s built on WebContainers technology by the StackBlitz team and supports frameworks including React, Next.js, Astro, and Vite.
How much does Bolt.new cost? Bolt.new offers a free tier with 1M tokens. Paid plans start at $20/month with 10M tokens, with usage-based pricing beyond that [5].
Bolt.new vs Cursor: which should I use? Bolt is browser-based and built for non-developers. Cursor is a desktop IDE designed for developers who want AI assistance in their existing coding workflow. As Alexander put it: “We are easier to use than Cursor, especially for non-developer users.”
Bolt.new vs Lovable: what’s the difference? Both are browser-based and beginner-friendly. Alexander says Bolt is more enterprise-ready, with WebContainers architecture that keeps code local in the browser and a hands-on enterprise integration service for design systems and code requirements.
Bolt.new vs Replit: which is better? Bolt runs everything in-browser via WebContainers with no remote server. Replit uses cloud-based environments. Both target non-developers, but Bolt emphasizes zero-setup and enterprise readiness.
Can Bolt.new build production apps or just prototypes? Both. Alexander says depending on the use case, Bolt can get you “all the way across the line to an application that is deployed and used by your users and customers.”
What’s the fastest way to go from idea to working app? Type your idea into Bolt. Alexander demoed a fully functional app in under two minutes during our conversation.
How do teams validate product ideas before committing engineering resources? Use Bolt to build a working prototype in minutes instead of weeks. Product managers can test ideas with stakeholders before pulling engineers off the roadmap.
Is Bolt.new enterprise-ready? Bolt is actively deploying engineers into enterprise customers to integrate design systems and code requirements. Alexander expects a fully productized enterprise solution by 2026.
What hidden features does Bolt.new have? Visual inspector (lets you click and modify specific UI elements directly) and Plan mode (forces the AI to generate a reviewable plan before building). Alexander estimates only 5% of users know about these.
References
[1] Kyle Poyar, “How Bolt.new hit $40M ARR in 5 months,” Growth Unhinged, July 2025. https://www.growthunhinged.com/p/boltnew-growth-journey
[2] Sacra, “Bolt.new at $40M ARR,” February 2025. https://sacra.com/research/bolt-new-at-40m-arr/
[3] Anna Arteeva, “Choosing your AI prototyping stack: Lovable, v0, Bolt, Replit, Cursor, Magic Patterns compared,” Medium, January 2026. https://annaarteeva.medium.com/choosing-your-ai-prototyping-stack-lovable-v0-bolt-replit-cursor-magic-patterns-compared-9a5194f163e9
[4] PostHog Newsletter, “From $0 to $40M ARR: Inside the tech that powers Bolt.new.” https://newsletter.posthog.com/p/from-0-to-40m-arr-inside-the-tech
[5] Bolt.new User and Revenue Statistics, Shipper, January 2026. https://shipper.now/bolt-new-stats/
[6] Nucamp, “Top 10 AI Business Ideas You Can Start in 2026,” February 2026. https://www.nucamp.co/blog/top-10-ai-business-ideas-you-can-start-in-2026-low-cost-high-potential
[7] Humai Blog, “Vibe Coding Tools Comparison 2025,” December 2025. https://www.humai.blog/vibe-coding-tools-comparison-2025-cursor-vs-bolt-vs-lovable-vs-windsurf-vs-replit/