One AI Rulebook, Enterprise Agents at Scale, and Faster Local Clusters

🧩 The Gist

Washington moved to centralize AI governance, signaling a push for one national framework that overrides conflicting state rules. Enterprises are scaling agent workflows, with a major bank rolling out an internal platform used by tens of thousands of employees. Developer tooling took a step forward as macOS added RDMA over Thunderbolt for speedier multi‑machine AI work, and OpenAI aligned on portable “skills” that package repeatable tasks. Education efforts are also ramping up to demystify how AI works, and to clarify what it does not do.

🚀 Key Highlights

• The White House published a presidential action titled “Ensuring a National Policy Framework for Artificial Intelligence,” aimed at eliminating state law obstruction to a national AI policy.
• BNY is deploying OpenAI tech via its Eliza platform, with 20,000+ employees building AI agents to improve efficiency and client outcomes.
• Apple’s macOS 26.2 release notes reference RDMA over Thunderbolt, enabling faster small AI clusters across connected Macs.
• OpenAI has quietly adopted “skills” in ChatGPT and the Codex CLI, a folder based mechanism for packaged capabilities.
• In ChatGPT’s Code Interpreter, a /home/oai/skills directory is accessible by prompt, and was verified by zipping and inspecting its contents.
• OpenAI describes shipping Sora for Android in 28 days using Codex, citing AI assisted planning, translation, and parallel coding workflows.
• Raspberry Pi highlights classroom materials that use secondary school maths to show that AI systems do not think, supporting AI literacy.

🎯 Strategic Takeaways

• Policy and compliance
  • A single national framework reduces patchwork risk for companies planning AI deployments, and concentrates rulemaking at the federal level.
• Enterprise adoption
  • Internal agent platforms are moving from pilots to organization‑wide use, tying AI directly to efficiency and client facing outcomes.
• Developer platforms
  • OS level RDMA over Thunderbolt broadens accessible multi‑machine options for AI work without specialized datacenter gear.
  • Folder based “skills” suggest a portable, auditable way to share tasks and workflows across tools.
• Education
  • Clear, maths grounded explanations help set realistic expectations about AI capabilities and limits.

🧠 Worth Reading

• OpenAI are quietly adopting skills, now available in ChatGPT and Codex CLI
  Core idea: “Skills” are simple folders with a Markdown file and optional resources that tools can read, making capabilities portable and inspectable. In ChatGPT’s Code Interpreter, users can access a built in skills directory, which was zipped and explored to reveal instructions for tasks like PDF handling. Practical takeaway: teams can package repeatable workflows as skills for consistency and reuse across development and analysis environments.