Intel / article

sparqlite: SPARQL 1.1 inside SQLite, written in Rust

Source: simonwillison.net
Date: Published Oct 16, 2026

Why it matters

Demonstrates Claude Code's dynamic workflows on a hard systems task, with measured results (3.5x speedup, 636 of 639 W3C tests). Shows what this mode can deliver from a short prompt.

Key takeaways · AI-distilled

Simon Willison gave Claude Code for web in Ultracode mode, which generates JavaScript to coordinate networks of subagents, a one-line prompt containing a typo; it worked for over 12 hours and built a SPARQL 1.1 engine as a Rust SQLite extension.
The project's summary reports 3.5x speedups over SQL translation in count mode and 2.8x when returning full results, and 636 of 639 applicable W3C tests passing, from integer-encoded terms, covering indexes and a cost-based planner.
Willison writes that reviewing the generated code is not remotely feasible and even following its documentation takes significant work, and questions the value of a full implementation created on a whim without time invested to understand it.

Terms in this piece · Glossary

multi-agent — Using several AI agents on one problem — splitting work in parallel, checking each other, or filling different roles like planner and reviewer.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.

Read the source simonwillison.net