
An AI tool that turns text prompts into full-stack web and mobile apps in the browser.
Bolt gained an edge by adding its in-browser web-stack virtual machine to an AI coding copilot, and it went from 0 to $50 million in six months. It also won a seven-figure deal from Lovable after nobody there called back. Builders outgrew it on payments, finding it too opinionated and prone to bugs. Lesson: pair a technical edge with fast follow-up, and add planning before code.
From Zevi Arnovitz's advice on Bolt
Bolt won by adding its in-browser virtual machine to an AI coding copilot. The team worked in obscurity and open sourced a project, then combined their web-stack virtual machine that runs in the browser with an AI coding copilot and ended up with something better than anyone else had.
Mark Pincus was impressed enough by this story that he cold-emailed the founder.
At hypergrowth, deciding which inbound leads to answer becomes half the job. Bolt went from 0 to $50 million in six months, and half the work was choosing which leads to respond to.
It closed a seven-figure deal that would have gone to Lovable, because nobody at Lovable called the buyer back.
Bolt makes non-coders feel superpowered by making decisions for them. Zevi Arnovitz, a Meta PM who doesn't write code, started on Bolt. Like Lovable, Replit, Base44 and v0, it uses the same underlying models but adds a layer that picks things like sign-in and database for you.
That makes building easy but gives less control. He left when connecting payments got too hard and the tool felt too opinionated, moving to Cursor and Claude Code for full tool access and the latest model abilities. Quote: "I think that the main difference between all these tools is basically the harness."
Bolt codes immediately, so complex work needs a planning step first. Bolt's coding-agent system prompt starts writing code at once. That works early on but caused really gnarly bugs in his apps on complex work like payments and database changes.
His fix is to have the AI read the codebase, ask clarifying questions about scope, data model, UX and grading, and write a plan before any code. Quote: "planning is really important when you're implementing something technical"
Treat Bolt as one step on a path from chatbot to terminal. He recommends starting in a ChatGPT or Claude project, moving to Bolt or Lovable, then Cursor in light mode, and eventually the terminal, moving up only when you outgrow each tool.
Looking at code works as exposure therapy, and because code is just files, the same project carries across tools.
A free year of Bolt drew fraud rings to Lenny's Product Pass. Bolt was one of five AI tools, with Cursor, Lovable, Replit and v0, offered free for a year to Lenny's paid subscribers. Fraud rings, mostly in China, found exploits in the API and spread them through student networks.
Lenny worked with Stripe and Substack to shut them down while his engineer barely slept for a week. He calls it his most stressful business moment. Quote: "that was way too good an offer. People came for it, bad people."
Reached $20M ARR two months after its October launch and later passed $30M, with about a million monthly users and 67% non-developers. Built in 90 days on a team of 15 to 20, with a yearly goal of $100M ARR.
Bolt won by adding its in-browser virtual machine to an AI coding copilot. The team worked in obscurity and open sourced a project, then combined their web-stack virtual machine that runs in the browser with an AI coding copilot and ended up with something better than anyone else had.
Mark Pincus was impressed enough by this story that he cold-emailed the founder.
Its team worked in obscurity, open sourced a project, then added their web-stack virtual machine to an AI coding copilot and had something better than anyone. Pincus cold-emailed the founder, impressed by the story.
A free year of Bolt drew fraud rings to Lenny's Product Pass. Bolt was one of five AI tools, with Cursor, Lovable, Replit and v0, offered free for a year to Lenny's paid subscribers. Fraud rings, mostly in China, found exploits in the API and spread them through student networks.
Lenny worked with Stripe and Substack to shut them down while his engineer barely slept for a week. He calls it his most stressful business moment. Quote: "that was way too good an offer. People came for it, bad people."
One of five AI tools offered free for a year to Lenny's subscribers in the Product Pass.
Bolt makes non-coders feel superpowered by making decisions for them. Zevi Arnovitz, a Meta PM who doesn't write code, started on Bolt. Like Lovable, Replit, Base44 and v0, it uses the same underlying models but adds a layer that picks things like sign-in and database for you.
That makes building easy but gives less control. He left when connecting payments got too hard and the tool felt too opinionated, moving to Cursor and Claude Code for full tool access and the latest model abilities. Quote: "I think that the main difference between all these tools is basically the harness."
Bolt codes immediately, so complex work needs a planning step first. Bolt's coding-agent system prompt starts writing code at once. That works early on but caused really gnarly bugs in his apps on complex work like payments and database changes.
His fix is to have the AI read the codebase, ask clarifying questions about scope, data model, UX and grading, and write a plan before any code. Quote: "planning is really important when you're implementing something technical"
Treat Bolt as one step on a path from chatbot to terminal. He recommends starting in a ChatGPT or Claude project, moving to Bolt or Lovable, then Cursor in light mode, and eventually the terminal, moving up only when you outgrow each tool.
Looking at code works as exposure therapy, and because code is just files, the same project carries across tools.
His first tool, which made a non-coder feel superpowered. He outgrew it when connecting payments, since it starts writing code at once and felt too opinionated, causing gnarly bugs.
At hypergrowth, deciding which inbound leads to answer becomes half the job. Bolt went from 0 to $50 million in six months, and half the work was choosing which leads to respond to.
It closed a seven-figure deal that would have gone to Lovable, because nobody at Lovable called the buyer back.
It went from 0 to $50 million in six months, and half the job was choosing which leads to answer. It closed a seven-figure deal taken from Lovable because nobody there called back.
It launched in New Zealand or Australia with top design and speed, but cohorted retention never flattened, so it failed. Some technology was ported to the main app.
It relies on long, nuanced product prompts, an example of product-focused prompt engineering.
Tried to build before Claude 3.5 Sonnet made its product work. It is an example of Anthropic's models powering builder products.
Bolt writes new code from scratch, while Windsurf works in a million-line code base without breaking it.
Its builders spent years on a product that failed until Claude Sonnet 3.5 arrived and suddenly made it work.
Osika contrasts Lovable with Bolt as a rival tool, saying Lovable differs by packaging AI output for non-technical users.
Developer toolsA chip company whose GPUs and CUDA software power gaming, AI and accelerated computing.
Developer toolsAnthropic's agentic coding tool that lets Claude work on code end to end.
Developer toolsAn AI code editor that lets engineers build software with AI while staying in control.
Developer toolsA code hosting and collaboration platform for software developers, with free and paid plans.
Developer toolsAn AI app builder that lets people create software by chatting with an agent.
An app development company that builds apps for founders and clients.
Developer toolsA fast, well-designed project and issue tracking tool for software teams.
Developer toolsGoogle's backend platform for building apps.