Independent developer Simon Willison recently did something small but highly representative: after accumulating 244 AI-assisted development tools, he now has models read their code, automatically identify the JavaScript interfaces, software libraries, and network services they use, and generate tags for discovery. The number shows that AI has pushed much of people's work beyond writing a single tool and into maintaining hundreds of them at once.
What this is
Simon Willison originally wanted a better way to browse his collection of tools. Using his models' remaining quota, he asked AI to analyze each project with tree-sitter, a parsing tool that breaks source code into a structured syntax tree, then generate a tag cloud on his homepage by technical category.
The key is not the tag cloud itself, but the extraction of dependencies previously scattered throughout source code into searchable information. Finding out which interfaces a project used once required opening files one by one; now, users can narrow the field by category first. This is a lightweight approach to code-asset management, well suited to independent developers and small teams whose tool collections are growing rapidly.
Industry view
We believe this demonstrates a clear signal: the next stage of AI programming will not only improve generation speed; it will also take on code inventories, dependency mapping, and knowledge retrieval. As the barrier to development falls and tool collections multiply, the ability to find and understand old tools becomes a new problem.
The risks are equally clear. Automatic tagging depends on code parsing and model judgment, so misclassification can distort search results. The more tools AI generates, the more license vulnerabilities, duplicate functionality, and outdated dependencies can be amplified. So far, this approach has only been demonstrated in a personal project and remains some distance away from reliable enterprise-grade code-asset management. It is more a landmark worth testing than a mature solution.
Impact on regular people
For enterprise IT: Companies can adapt this approach to catalog internal scripts, automation tools, and legacy systems by capability and dependency. For teams with many tools but inadequate documentation, its value may be more immediate than generating yet another application.
For individual professionals: People who use AI to build tools will find it easier to accumulate personal productivity systems, but tool count does not equal effective output, and maintenance costs will rise with it. Scarcity will remain in the ability to select, verify, and continuously integrate tools.
For the consumer market: AI applications will move beyond one-off question-answering sessions toward long-term usage records. Users will expect products not only to remember conversations, but also to organize the content, code, and workflows they accumulate over time while providing explainable classification results.