4 agent projects go to the Agentic AI Foundation
Anthropic, OpenAI and Alibaba each gave a tool for AI agents to one neutral group, with no test results.
The Agentic AI Foundation (AAIF), a neutral group for open standards for AI agents, added 4 projects. Anthropic gave Agent Skills. OpenAI gave Open Responses. Alibaba gave OpenSandbox. A group that includes Hugging Face, Microsoft and Meta-PyTorch gave OpenEnv. The announcement gives no test results or user numbers. The foundation already holds MCP, goose and AGENTS.md.
What each one does
- Agent Skills: an open standard that gives agents more skills.
- Open Responses: one format for messages to many model providers.
- OpenEnv: a shared tool to make training worlds for agents.
- OpenSandbox: a safe space to operate agent code, with tools for Docker and Kubernetes.
The numbers
The announcement gives no numbers for use, speed or cost. Anthropic says that it started Agent Skills 1 year ago. It says that it gave MCP to the foundation less than 1 year ago.
What is not known
It is not known how many teams will change to these standards. Other reports say that many providers use Open Responses, but Anthropic and Google use different formats.
Insight: Rivals put their standards in one neutral place
Anthropic gave MCP and at this time Agent Skills. OpenAI gave Open Responses, and Alibaba gave OpenSandbox. But Anthropic and Google use different formats. This reads as a race to own the layer below agents. Each firm gives code to make the format of the firm the standard. Builders must find which format the foundation uses first.
Sources
Posted
