agentic engineering
Octobench Finds a 5-Task Harness Swing
The same GLM model solved 24 coding tasks in one harness and 19 in another. Builders should benchmark the whole system.
The Weighted Average
Same-model task gain from the Octomind harness +5 — Octobench Finds a 5-Task Harness Swing Mean Build-stage error reduction with skills 80% — Supabase Evals Turns Agent Support Into Tests India AI-app revenue held by ChatGPT and Claude 83% — India's AI App Revenue Concentrates at the Top Announced deal value versus SkyWater FY2025 revenue 4.07× — IonQ's $1.8B Foundry Bet Tests Neutrality Minnesota law signing to TRO denial 85 days — Minnesota's AI Image Law Survives xAI's First Bid
The Wire
policy geopolitics
The EU AI Office Adds 38 Enforcers
Europe added 38 AI enforcers to an office reported above 140 people. Model vendors should rehearse the evidence request now.
enterprise ai work
AI-Exposed Jobs Show a 6.7-Point Wage Gap
Apollo finds a 6.7-point wage-growth gap in AI-exposed jobs without significant job loss. Employers should audit careers, not only cuts.
Quoted
“A coding model never arrives alone; retrieval, memory, stopping rules, and tests walk into production with it.”
About the Paper
“I’m Stephen, a software engineer with over two decades of industry experience. The Weighted Average is a daily reading of the AI economy — the models, the money, and what they do to us.”