📡 Radar — 8 Sep 2026

The daily catch for 8 September — all days →

📡 8 Sep 2026

🛠️ AI Development

News
A tough new benchmark drops coding agents into real client engagements
A new benchmark drops a coding agent into a realistic client engagement and grades the delivered work: Claude Opus 5 under Claude Code passes 23.9%, an expert human 82.2% — a sobering gap.
via @dair_ai ↗
News
ETH Zürich: the students who leaned on AI most were worst at vibe coding
First preregistered study on who's good at vibe coding (CHI '26): CS background and writing skill predict success, while the heaviest AI users did worst — a counter to 'just prompt more'.
via @Suryanshti777 ↗

Article
What AI actually looks like inside five enterprise design teams
A field report from design teams at Atlassian, Stripe and Ramp: what separates real AI use is documented design systems and prototyping infrastructure — not the model — plus deliberate adoption work.
Post
A popular roundup of tools that turn visual direction into 'design markdown' for agents
A widely-shared thread collecting nine sites that emit structured 'design markdown' — style and design-system specs to feed AI coding agents for more consistent output. Practical, if listicle-shaped.
via @Hartdrawss ↗