Blog

Every claim I caught rotting last week was true when someone wrote it. That is what makes them dangerous — and what makes them predictable.

smolagents promises its agent logic fits in ~1,000 lines, and links to the file to prove it. The file is 1,813 lines. This is what happens when your pitch is...

Label rot is silent, structural, and more common than you think. Here’s how three triage passes cleaned up a schema classification test set — and why guardin...

On Signal, Noise, & Sustainability : The limits of overproducing

Your AI fears are valid—they’re just aimed at the wrong target. The real crisis isn’t replacement. It’s drowning in mediocrity we can’t measure.

From Hemingway Editor to textstat: a complete guide to readability tools for writers who want clearer prose and developers who need programmatic text analysis.

Research synthesis examining why context management is critical for AI coding agents, based on the ACE-FCA (Advanced Context Engineering for Coding Agents) f...

Traditional LLM evaluation focuses on final outputs. But AI agents take multi-step actions—and measuring just the result misses where things actually went wr...

A comprehensive guide to minimizing context usage, optimizing token consumption, and maximizing efficiency when working with Claude Code and the Claude API.

Or: graphs are cool, but it would be a whole lot cooler if they were DAGs

On Signal, Noise, & Sustainability : The limits of overclocking

How to make your Wix website run 75% faster on 60% less data and ~$400 less in hosting fees - instantly.

Very much a work in progress, started in 2017 and picked up again a few weeks ago.

A guide to updating Jekyll after 8 years - from 2017 setup to 2025 best practices

My Quick Style Reference for Jekyll/Markdown/Redcarpet