The agentic coding story keeps compounding: Anthropic says 80% of its own new production code is now Claude-authored while publishing research on recursive self-improvement, even as Uber slaps hard monthly spending caps on Claude Code and similar tools after costs ran away, and Spotify publishes its own playbook for scaling agent-driven development across teams. Security cuts both ways today - an AI tool dug up a two-year-old critical Redis bug, but researchers also chained Zapier sandbox primitives into a near-hijack and demonstrated a worm that could propagate across networked AI agents. On the product side, Google shipped Gemma 4 12B and the cartoon-generating Dreambeans app, Blackmagic released DaVinci Resolve 21 with new AI editing tools, and Elixir 1.20 quietly became a gradually typed language. Money continues flowing into agent infrastructure and frontier labs, with Morgan Stanley opening its wealth funnel to AI agents, Microsoft repackaging OpenClaw as an enterprise agent called Scout, and VoidZero (the JS tooling shop behind Vite's creator) joining Cloudflare. For engineers, the throughline is that agentic coding has moved from experiment to line-item cost and governance problem almost overnight.

• Anthropic says 80% of its new production code is now Claude-authored
• Uber caps AI coding tool spend at $1,500/employee/month after costs spiked
• AI security tool surfaces a critical, two-year-old Redis takeover bug
• Researchers chained Zapier sandbox flaws into a near full-platform hijack
• Google ships Gemma 4 12B, an encoder-free unified multimodal model