OpenAI acquired Glass Imaging for over $300M (WSJ via TechCrunch, Sept 14). The Los Altos startup—founded by ex-Apple Portrait Mode engineers—uses neural nets tuned to each phone camera to improve shots at capture time, fitting OpenAI’s rumored device plans after the $6.5B io deal.
Since Oct 29, the AI Bots managed ruleset on all plans takes a separate Log or Deny action for AI crawlers (training bots like GPTBot and ClaudeBot) and AI assistants (fetchers like ChatGPT User and PerplexityBot), set under Security Firewall Rules or with vercel firewall rules edit ai bots category ai crawlers action deny . Check bots.fyi first, since Applebot sits in the crawler bucket too…
Since Oct 29 (Agents SDK v0.32.0, Think v0.23.0), a folder with a SKILL.md dropped into /workspace/.agents/skills/ , an R2 bucket or code is picked up on the agent's next turn, checked against the open Agent Skills spec, and usable from Think, Pi Durable or the AI SDK. It's experimental and breaking: SkillRegistry , the agents:skills Vite import and agents/skills/compile are gone…
Since Oct 30, a new table in the public schema of any existing project is unreachable through supabase js, REST and GraphQL until you explicitly grant anon , authenticated or service role access, and a missing grant returns error 42501 with the exact GRANT to run. Existing tables keep their grants and RLS works as before, so if an AI tool creates your tables, have it add grants and RLS policies in the same migration…
Announced Oct 29 in waitlist beta, macOS Satellites boot a fresh cloud Mac with Xcode and iOS simulators in about 30 seconds, while Amp keeps editing on its Linux orb and streams the Mac's screen so you can watch or take over. Macs cost $2/hour (mac.medium) or $4/hour (mac.large) from your orb usage, and are deleted after 20 idle minutes…
In experimental preview since Oct 28, Fusion Standard in Copilot CLI, the Copilot app and VS Code can hand eligible tasks to a local model, with managed MAI Code 1.1 Flash download on Windows (NVIDIA GPU, 96 GB RAM) or any OS via a local OpenAI compatible server running MAI Code 1.1 Flash or Qwen3.8 Flash Next. Local calls don't use GitHub AI Credits, but routing still happens in the cloud, so sessions aren't offline…
Spotted something we missed? Start a thread.
