Multi-agent coding tools dominated today's launches — a clear sign the parallel AI workflow is becoming the default for developers.
Claude's security research capabilities are getting serious attention, Cursor is self-improving every 5 hours, and China is scaling humanoid robot production.
Nicolas Carlini (67K citations, Anthropic Research Scientist) demonstrated Claude finding a Linux kernel buffer overflow introduced in 2003 that was never discovered by humans, a blind SQL injection in Ghost (50K GitHub stars), and earning $3.7M exploiting smart contract vulnerabilities.
He says Claude is "a better security researcher than me" and expects LLMs to only get better. The Linux exploit is especially notable — buffer overflows are notoriously hard to find, even for experts.
Quick hits from the local LLM world.
Klein 9B's edit mode is finding practical uses, and a new reasoning LoRA is transforming LTX 2.3 video quality.
A local voice translator for Mac, IPTV management for the arr stack, and a couple of solid open-source tools.