Bolt.new built on Claude Sonnet and WebContainers

Bolt worked because a new model made one shot app generation good enough at the same moment StackBlitz already had the runtime to execute that app instantly in a browser tab. Claude 3.5 Sonnet shipped in June 2024, and Anthropic presented it as a major coding upgrade. StackBlitz had already spent years building WebContainers, which run Node.js inside the browser instead of on rented cloud machines. That combination turned a prompt into a live full stack app with much less waiting and much lower infrastructure cost than cloud first rivals.

• The hard part was not just writing code, it was making the code runnable right away. WebContainers boot full Node.js environments in milliseconds inside the browser, which let Bolt show the result, open a terminal, install packages, and preview the app without spinning up a separate server session for each user.
  2 anthropic 3 stackblitz 4 sacra 5 sacra
• That architecture changed the economics. Bolt says WebContainers remove a major cloud VM bill and pushed gross margins close to 70%, unusually high for an AI coding product where inference is already expensive. This is why Bolt could spend tokens on Claude for generation while keeping the rest of the editing environment cheap to operate.
  4 sacra 5 sacra 6 sacra
• Peers prove the market split. Lovable also uses frontier models for app generation, but orchestrates multiple models and prices around chats. Cursor and Copilot use a layered model stack inside a local or desktop IDE workflow. Bolt sat in a different lane, natural language app creation plus immediate in browser execution, which made it feel more like a finished product builder than an assistant inside an editor.
  5 sacra 6 sacra 7 sacra

Going forward, the same intersection is becoming a race to own more of the stack. As model quality rises, the durable edge shifts toward who can turn generated code into production ready software with the right context, design system inputs, and deployment path. Bolt started with model timing plus browser infrastructure. The next phase is turning that speed advantage into a sticky enterprise workflow.