Simon Willison 2/10/2026

Structured Context Engineering for File-Native Agentic Systems

Read Original

A new paper systematically studies context engineering for structured data in file-native agentic systems. It details 9,649 experiments using SQL generation as a proxy, testing 11 models (including frontier and open-source) across 4 file formats (YAML, Markdown, JSON, TOON) on schemas from 10 to 10,000 tables. Key findings highlight the superiority of frontier models, their benefit from filesystem context, and the 'grep tax' penalty for the token-efficient TOON format due to model unfamiliarity.

Structured Context Engineering for File-Native Agentic Systems

Comments

No comments yet

Be the first to share your thoughts!

Browser Extension

Get instant access to AllDevBlogs from your browser

Top of the Week