Claude Artifacts is described in Not all AI-assisted programming is vibe coding (but vibe coding rocks) as one of the first widely available platforms for Vibe Coding — building software by prompting a language model rather than by writing and reviewing code. What distinguishes it in that account is not its generation ability but its containment: the code it produces runs inside a sandbox that limits what a program its author has not read can do.
Simon Willison singles the sandboxing approach out as fantastic and treats it as the model for how vibe coding is made safe for people new to building software. The same post is candid that the containment is a real trade-off rather than a free one, and that the restrictions rule out whole categories of project.
Capabilities
Generated code is restricted to running in a locked-down iframe, can load only approved libraries, and cannot make network requests to other sites. Willison’s assessment is that this combination makes it very difficult for someone to mess up and cause harm elsewhere with a project — the sandbox absorbs the accidents that unreviewed code would otherwise cause.
The same restrictions bound what can be built. A project running under them cannot reach data from external APIs, and cannot run its own prompts against an LLM. Willison names both limits explicitly while still endorsing the design.
Adoption & Ecosystem
The post positions Claude Artifacts at one end of a spectrum of vibe coding tools, with tools originally aimed at professional developers — Cursor is the example given — at the other, carrying far fewer safety rails. Willison presents this contrast as an open design problem rather than a settled ranking, saying he hopes to see a wide proliferation of tooling that helps people build their own custom tools both productively and safely.