2026-11-11 · ← News
Simon Willison used tree-sitter to map 244 AI-assisted tools
Simon Willison added code-derived tags to a collection of 244 mostly AI-assisted web tools using tree-sitter. The small site upgrade illustrates a useful pattern: AI can write software, but deterministic parsers are better for inventorying what it actually produced.
The image could not be loaded.
Simon Willison has made his collection of 244 web tools browsable through tags derived from their source code. He used remaining Fable 5.2 credits to build a tree-sitter based analysis that identifies JavaScript APIs, libraries and web API hosts used by each tool.
The tags now describe actual code rather than a hand-maintained catalogue
The tools.simonwillison.net homepage now offers tag clouds for three categories: JavaScript APIs, loaded libraries and web API hosts. Selecting a tag lists the tools that use that component.
Willison describes the collection as an experiment in prompt-driven development for low-stakes projects. At 244 entries, however, a plain list is no longer enough. Automated analysis extracts metadata from each small project without requiring manual descriptions to be rewritten.
Generated software needs a machine-readable inventory before it outgrows its author
Vibe coding makes small applications cheaper to create and consequently more expensive to navigate and maintain later. For one person, 244 tools already form a portfolio whose libraries, browser APIs and external services are difficult to remember.
Teams that let agents create internal utilities quickly will face the same problem. A catalogue derived from code can support discovery, duplicate consolidation and an initial dependency overview. The return to a conventional parser is the important part: a model can prepare the solution while a predictable tool verifies the code’s structure.
A tag cloud is still neither a security audit nor a complete dependency list
The short announcement does not describe extraction accuracy, handling of dynamically loaded code or whether the analysis runs automatically after every change. A tag only says that an element appeared in the code. It does not say whether that element is current, vulnerable or reachable in normal use.
The collection is deliberately a low-stakes experiment, so it cannot be copied directly into an enterprise software process. There, the inventory would also need versions, licences, code provenance and clear ownership for fixes.
Automatic refreshes and exposed duplication will determine practical value
The tags will prove useful if they update with every new tool rather than remaining a one-off map. Another signal will be whether the catalogue exposes repeated solutions, questionable external services or project groups worth consolidating.
For teams, the principle matters more than the cloud itself. Once agents produce dozens of applications, machine-readable inventory becomes an operational requirement rather than homepage decoration.
Lilith's verdict
AI built 244 small stalls and tree-sitter finally drew a map of the market. Keep letting agents build without an inventory and soon you will not know what you own or who is collecting tolls inside it.