OpenAI Skills, 10K Robots, and Runway's World Model

The AI agent ecosystem reached a new inflection point this week as OpenAI quietly adopted Anthropic’s skills framework across ChatGPT and Codex CLI, signaling convergence toward interoperable agent standards. Meanwhile, 1X secured a landmark deal to deploy up to 10,000 humanoid robots to industrial sites, Runway unveiled its General World Model for real-time simulation, and Google brought live translation to any pair of headphones. These developments underscore how quickly AI infrastructure is maturing from research curiosity to production-ready deployment.

OpenAI Adopts Skills Framework for ChatGPT and Codex CLI

OpenAI has quietly rolled out a modular skills system across ChatGPT and Codex CLI, mirroring the approach Anthropic pioneered in October. Developers discovered a /home/oai/skills directory in ChatGPT’s Code Interpreter containing skill.md files for PDFs, spreadsheets, and documents. The Codex CLI implementation uses a similar structure at ~/.codex/skills, allowing developers to define custom capabilities through Markdown files. By cloning the competitor’s standard, OpenAI effectively validates MCP as the industry default for agent definitions, moving the ecosystem toward interoperable, local-first agent skills.

1X Partners with EQT to Deploy 10,000 Humanoid Robots

Norwegian robotics startup 1X announced a strategic partnership with EQT to deploy up to 10,000 Neo humanoid robots across the private equity firm’s global portfolio between 2026 and 2030. The deal pivots the $20,000 consumer robot toward manufacturing, warehousing, and logistics applications across EQT’s 300+ portfolio companies. Pilots begin in the United States in 2026 with potential expansion into Europe and Asia. The agreement positions 1X against Tesla, Figure AI, and Boston Dynamics in the race to commercialize humanoid robotics at scale.

Runway Unveils General World Model GWM-1

Runway introduced GWM-1, a General World Model designed to simulate interactive environments in real time with understanding of physics, geometry, lighting, and cause-effect relationships. Built on Gen-4.5, the autoregressive model generates frame-by-frame at 24 fps and 720p resolution, supporting interactions lasting several minutes. The release includes three specialized branches: GWM Worlds for explorable environments, GWM Avatars for realistic speaking characters, and GWM Robotics for synthetic robot training data. Runway plans to merge these capabilities into a unified model and offer GWM-Robotics via SDK.

Google Translate Adds Real-Time Headphone Translations

Google rolled out a beta experience enabling real-time translations directly through any pair of headphones, powered by Gemini’s native speech-to-speech capabilities. The feature preserves each speaker’s tone, emphasis, and cadence while supporting over 70 languages on Android in the U.S., Mexico, and India. Users can access the capability through the Translate app’s new “Live translate” button. Google plans to expand to iOS and additional countries in 2026.

World Launches Super App with Encrypted Messaging and Crypto Payments

Sam Altman’s identity verification project World released major super app updates including end-to-end encrypted messaging and Venmo-style crypto payments. World Chat uses color-coded bubbles to indicate iris-scan verification status, with encryption described as equivalent to Signal. The app now supports USDC, EURC, wrapped Bitcoin, Ethereum, and 100+ cryptocurrency tokens. Despite targeting 1 billion users, fewer than 20 million people have completed verification through the Orb biometric system.

Oracle Delays OpenAI Data Centers to 2028

Oracle pushed back completion dates for several OpenAI data centers from 2027 to 2028 due to labor and material shortages. The delays affect sites in Texas and Ohio that are part of the $300 billion Stargate AI infrastructure program jointly announced with SoftBank. Oracle maintains that all contractual milestones remain on track, noting 96,000 Nvidia Grace Blackwell GB200 chips have already been delivered to its Abilene, Texas supercluster. The delays highlight broader bottlenecks in AI infrastructure buildouts as demand continues to outpace supply.

This week’s announcements reveal AI infrastructure maturing across multiple fronts: agent interoperability standards gaining consensus, humanoid robotics moving from lab demos to enterprise contracts, and world models enabling real-time simulation. The Oracle delays serve as a reminder that even trillion-dollar commitments face physical constraints. Watch for skills adoption to accelerate as developers embrace the emerging cross-platform standard.