Tools as Data, Not Code
Why retrieval-based AI tool use requires treating tools as data, not code—and what that makes possible.
Page 2 of 2
Why retrieval-based AI tool use requires treating tools as data, not code—and what that makes possible.
Extracting minimal schema subgraphs from large API specs sounds simple. It isn't. Here's what we learned building compiler-grade transformation infrastructure.
JSON Schema processors typically implement keywords as cases in a switch statement. We made each keyword a first-class citizen. Here's why that matters.
A different approach to document processing: what if the parsed JSON already is the intermediate representation?