An open source development environment called Bolt.new, currently in beta, combines AI and Web Containers, running Node.js inside the browser, with documentation that stresses getting the AI prompts right as much as understanding web technology.
Described as “AI-powered full-stack web development in the browser,” StackBlitz Bolt.new is in part open source on GitHub – though the published code is aimed at “developers interested in building their own AI-powered development tools” rather than being a ready to run full deployment.
Bolt.new presents a prompt as the centerpiece when starting a new application – though it is also possible to select from a StackBlitz template such as a blog using Astro, or a documentation site using Vitepress. The documentation emphasizes using AI chat to get started, with remarks about being as precise as possible in the initial instructions “to ensure Bolt scaffolds the project accordingly.” There is also a note about combining simple instructions into a single message to save time and AI token consumption.
The pricing model is based on token consumption – $50 per month buys up to 26 million tokens. If the tokens run out, the AI will stop working until more are purchased.
The Bolt environment also includes a browser-based editor – based on the same technology as Visual Studio Code – so the matter of how much to trust the AI, and how much to write code manually, is a developer choice. The browser-based Bolt IDE is powered by WebContainers, allowing tools such as Node.js, npm and yarn to run in the browser.
The environment uses Anthropic for the AI and there is also an emphasis on the StackBlitz-sponsored Vite JavaScript build tool. Netlify is also involved, as the default deployment target for a Bolt application, though according to a thread on X CloudFlare will be added as an option.
According to StackBlitz, Bolt will generate high quality code thanks to “instrumented & integrated AI at every layer of WebContainers, making Bolt aware of errors and able to solve them for you.” It will take more than that to convince developers that AI-powered Bolt will create reliable and maintainable applications of any complexity. AI code assistants are more often used for smaller pieces of code, where the developer knows exactly what is required but uses AI to avoid the tedium of writing again what has been done a thousand times before.
It seems that another key factor in the quality of a Bolt application is the ability of the developer to use the right AI prompts – and that efficiency may be measured as much by the right prompts in the right sequence, as by traditional coding skills.
A developer going by the handle “mangonaise” lamented on Hacker News: “It’s so weird to me that this is how we’re programming user-facing systems now.”