📰 News
-
Apple sues OpenAI over alleged trade-secret theft — Complaint says hardware chief Tang Tan (24 years at Apple) used Apple project codenames while recruiting, told candidates to bring components to interviews, and coached leavers on dodging security. Another ex-Apple engineer allegedly kept a company laptop and confidential docs on unreleased tech. Timing matters: OpenAI is widely rumored to be building hardware that could compete with the iPhone. OpenAI says it has “no interest in other companies’ trade secrets.” → TechCrunch
-
Fidji Simo steps down as OpenAI’s No. 2 — Medical leave ran longer than expected; she leaves the Applications CEO seat for a part-time advisory role. She owned consumer growth and was often floated as a post-IPO successor — a real gap for Altman. He posted that the news “sucks.” → TechCrunch
-
Claude Code ships a built-in browser — Claude can open docs and issue trackers inside the app, click, and type. Extra classifiers gate write actions; no purchases, account creation, or CAPTCHA bypass without consent. Orgs can allowlist sites or turn tools off. → The Decoder
-
Claude Cowork’s real job: boring office glue work — Anthropic’s look at 1.2M sessions: ~33% business process/ops, ~16% content/copywriting. Coding is only 8.7% (devs use Claude Code). Reports, checklists, and decks — the work around the work. → The Decoder
-
Meta kills Muse Image’s no-consent Instagram face gen — @-mention a public account and generate AI photos by default; opt-out only. After backlash Meta said it “missed the mark” and pulled it within days. → The Decoder
-
S&P calls OpenAI a “key credit risk” for Oracle — Oracle cut BBB → BBB- (one notch above junk). AI capex may hit $95B by 2027; OpenAI is ~half of $638B in contractual obligations. Collapse would leave empty data centers. → The Decoder
-
Altman is now “pretty sure” AI is net job-creating — A pivot from mass-layoff warnings: “This is not what I expected.” Broader productivity and labor stats still don’t show a clear AI shock. → The Decoder
-
LinkedIn leads long-form AI slop — Pangram: ~1 in 4 posts over 250 words is AI-flagged; LinkedIn long-form hits 41% and drives most detected AI content despite being a third of the sample. Substack ~10%. → The Decoder
🔥 GitHub Trending
-
davila7/claude-code-templates — Ready-made Claude Code agents, slash commands, MCPs, and hooks. Install stacks with one
npx. -
google-labs-code/stitch-skills — Agent skills for Google Stitch (design/build). Works with Codex, Claude Code, Cursor.
-
openai/plugins — Curated Codex plugin examples (Figma, Notion, iOS/web apps) with skills, MCP, and agents.
-
wonderwhy-er/DesktopCommanderMCP — Terminal, file, and process control over MCP — including Remote MCP from ChatGPT/Claude web.
▶️ YouTube
- Minecraft Was Missing One Brilliant Idea — Two Minute Papers | Terrain Diffusion: generating Minecraft-style worlds with diffusion models, in a few tight minutes.
💬 Community
-
Fable gets another bump — Simon Willison | Anthropic extends Claude Fable 5 on paid plans (and higher Claude Code limits) through July 19. He argues the access uncertainty pushes users to OpenAI.
-
Tau: An Educational Coding Agent — Lobsters | A coding agent built to teach how agents work — not just hand you the answer.
-
A global workspace in language models — Lobsters | Anthropic research: hints of Global Workspace Theory-like structure inside LLMs.
-
A Prolog library for interfacing with LLMs — Lobsters | Classic logic programming meets generative models.
-
AI Surveillance and Social Progress — Lobsters | Bruce Schneier on AI surveillance and whether it can coexist with social progress.