Simon Willison ran an experiment with Claude Code to build a SQLite extension that executes SPARQL 1.1 queries directly on database rows. The project, named sparqlite, was written in Rust and generated by an AI model using dynamic workflows known as Ultracode. The process took over twelve hours to complete after an initial prompt requested the implementation.
Willison noted that reviewing the resulting code is not remotely feasible and following the documentation requires a significant investment of time. He is unsure of the project’s value because he created it on a whim and cannot guarantee he will understand its capabilities. The experiment highlights the gap between generative AI output and the manual effort needed to maintain complex systems.
* The generated code includes a tutorial for installing and running the extension.
* A script was created to populate the database with realistic example data.
* Documentation covers both the SQLite extension usage and SPARQL fundamentals.
