The most important announcement at this week's AGNTCon in San Jose wasn't a model launch. It was a surrender — four of them, delivered in a single announcement from the Agentic AI Foundation: Anthropic is donating Agent Skills, OpenAI is donating Open Responses, Hugging Face is donating OpenEnv, and Alibaba is donating OpenSandbox.
Read the room for a second. OpenAI and Anthropic compete for the same enterprise contracts, the same talent, and the same compute — and both just handed key pieces of their agent plumbing to the same neutral body. The foundation went from three launch projects in December 2025 to ten in a year, and the four newcomers cover the whole stack: how agents gain skills, how apps talk to models across providers, how agents are trained, and where they run safely.
The give-ups are strategic, not generous. Open Responses is based on OpenAI's own Responses API — and standardizing the interface is how you make the competition happen at the model layer, where OpenAI thinks it wins, instead of the plumbing layer, where nobody does. Databricks Model Serving and Vercel's AI Gateway already expose Open Responses endpoints, so this isn't hypothetical; it's ratifying what's already shipping. And Agent Skills, which Anthropic launched a year ago, has over 70 agent platforms supporting it and roughly 280 million installs, according to the foundation — the open format already won, so the donation just makes the victory official.
That's also why the real story is governance. "No single company controlling the outcome" is the line AAIF CTO Manik Surtani keeps repeating, and it doubles as antitrust insurance: when four rivals build on the same specs through a Linux Foundation-style body, nobody can accuse any of them of locking up the agent era. It's the same playbook that made Kubernetes and the Cloud Native Computing Foundation work — commoditize the infrastructure, compete on everything above it.
What to watch: adoption. Standards bodies are where good specs go to get debated, and MCP's own success (10,000+ servers, donated to AAIF less than a year ago) sets a high bar. If Agent Skills and Open Responses get the same pickup, the era of "works on one lab's stack" agents ends fast — and the next war is purely about whose models the standardized pipes connect to. As Alibaba's Xiaobin Xu put it, the point is that no single company defines the future. The race now is to be the model behind everyone else's standard.
— via Agentic AI Foundation