Writing

Thinking on context engineering.

Articles on AI agent architecture, token optimization, and building production AI systems.

NewToken DensityProgramming LanguagesContext Engineering

Your Programming Language Is an LLM Bottleneck

Python fits up to 90% more logic into an LLM context window than C#. 600+ identical algorithms tokenized across 6 languages to measure the hidden cost of verbose syntax.

Read article

Mar 1, 202612 min read

003

RLM ArchitectureLong-Context AIContext Engineering

Recursive Language Models: Beyond the Context Window

MIT's RLMs replace soft attention with hard programmatic search — cutting inference costs 40-64% while outperforming traditional LLMs on complex long-context tasks.

Read article

Mar 1, 20267 min read

002

TOON FormatAI AgentsContext Engineering

Half Your Tokens Are Wasted on JSON. TOON Fixes It.

JSON spends roughly half its tokens on structural noise. TOON eliminates that overhead with a single middleware layer — cutting token usage ~50% for AI agents consuming REST APIs.

Read article

Feb 25, 202614 min read

001