Developers working to integrate AI agents across different systems face persistent challenges: entrapment with single vendors, incompatible protocols, and the loss of context when moving work between frameworks. The Agentic AI Foundation, established by the Linux Foundation, aims to address these friction points by providing vendor-neutral infrastructure for agent development.

Since its December 2025 launch with three initial projects including the Model Context Protocol (MCP), AAIF has expanded significantly. On October 22, Anthropic and OpenAI each contributed new projects to the foundation, joined by submissions from Hugging Face and Alibaba, bringing the total to 10 hosted projects.

Manik Surtani, AAIF's co-founder and CTO, discussed the foundation's growth at AGNTCon + MCPCon in San Jose. He emphasized that while the underlying needs these projects address have long existed, the shift toward vendor-neutral governance represents a meaningful change. "I'm really happy [Agent Skills] is here. And Open Responses as well. I mean, having a standardized way to speak to a model is super important. And I'm glad to see that it's now happening in a way that is vendor-neutral, in a neutral governance and everything else."

Four major projects join the foundation

  • Agent Skills (Anthropic): An open format for packaging instructions, scripts, and reference materials that agents can load when executing specific tasks, enabling reuse across compatible agents.
  • Open Responses (OpenAI): An open specification derived from OpenAI's Responses API that standardizes how applications communicate with language models and receive results, reducing translation overhead when switching between model providers.
  • OpenEnv (Hugging Face): A framework for constructing and distributing reinforcement learning environments where agents train by performing tasks and receiving feedback.
  • OpenSandbox (Alibaba): A platform for executing agent-generated code and other operations in isolated environments, keeping that execution separate from the host system.

The convergence of major AI companies around a neutral foundation reflects developer demand that existed from the foundation's inception. Surtani noted that "More than faster, I think people are going to be able to build on top of things that they can trust more." This lesson emerged from the early cloud-native era, when enterprises learned the risks of building critical infrastructure on proprietary technology they could not influence.

Addressing agent memory and portability challenges

Surtani identified agent memory and context management as immediate priorities. Agents require access to information about prior work and current state to continue operating effectively. A harness—the software layer surrounding a model that orchestrates its tools and workflow—must carry this context when transferring work between different models or harnesses. Currently, this remains difficult, and Surtani advocates for teams to share context across agents and people collaborating on the same task.

Beyond context portability, Surtani stressed that engineering teams should build for model independence now through abstractions, proxies, and standardized interfaces. This approach prevents applications from becoming tightly coupled to a single model provider. A proxy or router positioned between the application and the model allows teams to swap the underlying model without disrupting the application layer. While not a complete solution, this pattern handles much of the complexity.

The agentic AI stack differs fundamentally from the cloud-native stack governed by the Cloud Native Computing Foundation (CNCF), which Surtani described as already well-established. The agentic AI stack remains inherently unstable. Surtani advised developers: "My advice to people is just be ready for change. Change is going to happen. Now, the good thing in all of this is that change is actually easy as well. With agents actually building themselves and writing code, it's not going to be that expensive or that hard to migrate from one thing to the other."

Sandbox phase for emerging projects

AAIF has attracted more than 300 member organizations within 10 months, spanning fintech companies and non-technology institutions like the Gates Foundation. The foundation recognized that its initial governance model, adapted from the mature CNCF template, created barriers too high for a rapidly evolving space.

To address this, AAIF introduced a new Sandbox lifecycle phase on September 1. This maturity category—distinct from the isolated execution environments provided by OpenSandbox—allows early-stage projects with strong community momentum to develop under neutral governance before reaching widespread enterprise deployment.

Surtani's primary message to developers: pressure your vendors to adopt open standards, publish transparent specifications, and commit to long-term stability.