Braid

VS Code Walks It Back, CAISI Signs Three Labs, and the Frontier Gap Compresses to Ten Weeks


Listen Later

Microsoft reverses the Co-Authored-by Copilot default it shipped last week, and that turns out to be one of three pieces of governance news today — alongside CAISI signing pre-deployment testing agreements with Google DeepMind, Microsoft, and xAI, and DeepMind's London staff voting 98% to unionize over military contracts. Then we go where the actual code lives: DeepSeek V4 Pro matching GPT-5.2 ten weeks later at one-seventeenth the price, a Qwen3.6 27B FP8 recipe that fits 200K tokens of unquantized KV cache on a single 48GB card, and a paper called AgentFloor that gives the small-model-routing intuition a benchmark to point at. Plus the tool-use tax, Chrome's silent four-gigabyte install, vibevoice.cpp, the Opus 4.7 complaint thread, and a B2B operator who replaced three vendors with a single Claude skill.

  • VS Code reverts the Copilot co-author default
  • CAISI signs Google DeepMind, Microsoft, and xAI
  • DeepMind workers vote to unionize
  • DeepSeek V4 Pro on FoodTruck Bench
  • Qwen3.6 27B FP8 with 200K BF16 KV on a single 48GB card
  • AgentFloor: how far up the tool-use ladder small open-weight models can go
  • The tool-use tax in LLM agents
  • Chrome silently installs a 4GB Gemini Nano
  • When everyone has AI and the company still learns nothing
  • vibevoice.cpp ports Microsoft VibeVoice to ggml
  • The Opus 4.7 regression thread
  • Replacing a 5-step lead enrichment chain with a Claude skill
  • ...more
    View all episodesView all episodes
    Download on the App Store

    BraidBy Lenar Kess · Damra Vol