Eric Simons and Bolt.new: Building Full-Stack Apps Inside the Browser

WebContainers made Bolt possible before Bolt existed

StackBlitz introduced WebContainers in 2021 as a browser-based environment capable of running Node.js, package managers, terminals and web servers with WebAssembly and modern browser APIs. CEO Eric Simons framed the technology as a way to build the web with the web, years before generative AI made it obvious that such an environment could also become a sandbox for an AI developer. ^[1] This episode matters historically because it shows how AI coding moved beyond isolated text generation and into a software-engineering workflow with context, tools, execution, or deployment around the model. The surrounding system shaped what users could delegate just as strongly as the underlying language model did.

The browser became the sandbox

Running the development stack in the browser reduced dependence on remote machines and made reset, preview and sharing immediate. The practical consequence was a shorter path between natural-language intent and a change that could be inspected in a real development environment, which became one of the defining characteristics of the vibe-coding wave.

StackBlitz launched Bolt at ViteConf 2024

StackBlitz’s ViteConf 2024 recap records the October announcement of Bolt, combining AI with WebContainers so users could prompt full-stack Vite applications into existence. A later company post explicitly identifies ViteConf 24 as Bolt’s launch moment and describes the rapid community uptake that followed. ^[2] This episode matters historically because it shows how AI coding moved beyond isolated text generation and into a software-engineering workflow with context, tools, execution, or deployment around the model. The surrounding system shaped what users could delegate just as strongly as the underlying language model did.

The launch joined two existing StackBlitz bets

StackBlitz had already invested in WebContainers and the Vite ecosystem, so Bolt could combine mature browser infrastructure with rapidly improving language models rather than build every layer at once. The practical consequence was a shorter path between natural-language intent and a change that could be inspected in a real development environment, which became one of the defining characteristics of the vibe-coding wave.

The AI received control of a real development environment

The public Bolt repository explains that the AI can control the filesystem, Node server, package manager, terminal and browser console. That is a crucial distinction from products that only emit source code: Bolt can install a dependency, observe an error, modify configuration and rerun the application inside the same browser session. ^[4] This episode matters historically because it shows how AI coding moved beyond isolated text generation and into a software-engineering workflow with context, tools, execution, or deployment around the model. The surrounding system shaped what users could delegate just as strongly as the underlying language model did.

Environment control enabled iteration

An agent that can inspect console output and terminal failures can correct its own work. That feedback loop is fundamentally different from one-shot source generation. The practical consequence was a shorter path between natural-language intent and a change that could be inspected in a real development environment, which became one of the defining characteristics of the vibe-coding wave.

Bolt reused the ordinary JavaScript ecosystem

Because WebContainers run ordinary Node.js projects, Bolt could use familiar frameworks and npm packages instead of inventing a proprietary runtime. This kept the generated output connected to conventional source code and let experienced developers inspect, edit, export or continue the project with familiar tools when natural-language prompting was no longer precise enough. ^[1] This episode matters historically because it shows how AI coding moved beyond isolated text generation and into a software-engineering workflow with context, tools, execution, or deployment around the model. The surrounding system shaped what users could delegate just as strongly as the underlying language model did.

Generated output remained normal web code

Developers could begin with natural language and still descend into React, configuration files and package dependencies, preserving a path from vibe coding to conventional engineering. The practical consequence was a shorter path between natural-language intent and a change that could be inspected in a real development environment, which became one of the defining characteristics of the vibe-coding wave.

The open-source Bolt codebase encouraged derivatives

StackBlitz published an open-source Bolt codebase and template for building AI applications on top of WebContainers. That release encouraged forks and experiments around local models, alternative prompting and new interfaces, turning Bolt into both a commercial product and a reference architecture for browser-native AI development. ^[4] This episode matters historically because it shows how AI coding moved beyond isolated text generation and into a software-engineering workflow with context, tools, execution, or deployment around the model. The surrounding system shaped what users could delegate just as strongly as the underlying language model did.

Browser execution shortened the prompt-to-feedback loop

The browser execution model removed much of the setup delay that normally separates an idea from a runnable project. Users did not need to install Node, select a package manager or configure a remote VM before the first prompt could produce a live preview, making experimentation cheap enough for non-specialists to treat software creation more like sketching. ^[5] This episode matters historically because it shows how AI coding moved beyond isolated text generation and into a software-engineering workflow with context, tools, execution, or deployment around the model. The surrounding system shaped what users could delegate just as strongly as the underlying language model did.

Bolt blurred the boundaries among IDE, hosting and app builder

Bolt compressed functions that had historically lived in separate tools: code generation, source editing, package installation, terminal execution, browser preview and deployment. That integrated loop helped define the vibe-coding builder as something different from an IDE plugin, because the environment could carry a project from prompt to running application inside one tab. ^[5] This episode matters historically because it shows how AI coding moved beyond isolated text generation and into a software-engineering workflow with context, tools, execution, or deployment around the model. The surrounding system shaped what users could delegate just as strongly as the underlying language model did.

Why Bolt.new belongs in the history of vibe coding

Bolt.new belongs in the history of vibe coding because it demonstrated that the execution environment can be as important as the model. WebContainers gave the agent a place to act and observe, and Bolt turned that infrastructure into a direct natural-language path from software idea to executable full-stack web application. ^[3] This episode matters historically because it shows how AI coding moved beyond isolated text generation and into a software-engineering workflow with context, tools, execution, or deployment around the model. The surrounding system shaped what users could delegate just as strongly as the underlying language model did.

Bolt also demonstrated how earlier infrastructure inventions can acquire new significance when AI capability changes. WebContainers were originally justified as a faster and more secure development environment for humans. Once models became capable of multi-step coding, the same sandbox became a ready-made computer for an agent. The historical lesson is that agentic products often depend on infrastructure that predates the agent wave.

The product’s browser-native design also affected collaboration. A generated application could be shared through a URL and opened without reproducing a local environment, reducing the friction of showing a prototype to a teammate or customer. That made vibe-coded software more social and iterative: the artifact was live and inspectable immediately rather than trapped in one developer’s machine.

RESEARCH / PROVENANCE

Works Cited

5 SOURCES

1. 01
   StackBlitz — Introducing WebContainers ↗blog.stackblitz.com
2. 02
   StackBlitz — ViteConf 2024 Recap ↗blog.stackblitz.com
3. 03
   StackBlitz — Bolt 100K Open Source Fund ↗blog.stackblitz.com
4. 04
   StackBlitz — bolt.new Repository ↗github.com
5. 05
   StackBlitz — Product Overview ↗stackblitz.com

CodeHistory is a living archive. Citations document the evidence used for this edition; later evidence may refine the account.