📡 Radar — 22 Jul 2026

The daily catch for 22 July — all days →

📡 22 Jul 2026

🛠️ AI Development

Post
Why benchmarks aren't enough to make coding agents good
Two engineers go deep on how SWE-bench and terminal-bench actually work, why they can't penalise an agent for writing slop, and the harder benchmarks now emerging from Cognition and others.
Post
Matt Pocock's 22 open-source AI skills for disciplined agent workflows
The skills Pocock uses daily to stop Claude Code and Codex from vibe-coding — turning coding agents into a requirement-first engineering process rather than one-shot generation.
Post
How 2,853 repos actually configure agentic coding tools
A paper surveying eight configuration mechanisms for agentic coding tools across 2,853 GitHub repos — and finding that one 'harness engineering' mechanism carries almost all the weight.

Post
A 'product brain' template for giving AI reliable design context
A concrete look at encoding business rules, spacing and taste into a reusable template, so AI design tools work from real product context instead of guessing.