TL;DR We added architectural context to AI coding agents via MCP and tested on SWE-bench...
How a $0.02/Call Model Scored 78.2% on SWE-bench Verified — Beating Every Model on the Leaderboard
Related
The Boring AI Is the Right AI
At the AI Engineer Summit 2025 in New York, the mantra that got repeated from stage after stage was...
Ten MCP servers I shipped this year. I use three.
An honest ranking of the ten MCP servers I built. Which three earn their slot in my config and which seven sit idle. The pattern is uncomfortable for the MCP hype cycle.
Five problems every agent loop has. No framework needed.
A short field guide. Five failure modes you will hit, the smallest library that fixes each, and the case against agent frameworks.