On 22 October 2026, four companies that compete fiercely in the AI agent market did something that signals where the enterprise stack is heading: they each contributed a project to a shared, vendor-neutral foundation. Anthropic handed over Agent Skills. OpenAI contributed Open Responses. Hugging Face brought OpenEnv. Alibaba submitted OpenSandbox. All four now sit inside the Agentic AI Foundation (AAIF), hosted by the Linux Foundation, under open governance that no single company controls.

For enterprise buyers, this is not an abstract standards story. It is a signal about interoperability, lock-in risk and the maturity of the agentic AI supply chain. When competitors agree on shared infrastructure, procurement decisions get simpler, integration costs fall and the platforms that adopt those standards become safer bets.

What was announced

The AAIF launched in December 2025 with three projects: the Model Context Protocol (MCP), goose and AGENTS.md. With the four additions announced at AGNTCon + MCPCon North America, it now hosts 10 projects in its first year. Each new project targets a different layer of the agentic stack.

Agent Skills (Anthropic). An open format for giving agents new capabilities and domain expertise. Skills package instructions, scripts and resources in a reusable format that agents can load on demand. The practical effect: a skill built for one platform can now be structured in a way that another platform can consume. For enterprises building custom agents, this means domain expertise does not need to be rebuilt every time the underlying agent platform changes.

Open Responses (OpenAI). An open specification for building multi-provider, interoperable large language model interfaces, based on the OpenAI Responses API. It defines a shared schema, consistent streaming events and extensible tooling. Developers can describe requests and outputs once and run them across model providers with minimal translation. For enterprise architecture teams, this means the application layer does not need to be rewritten when switching between model providers.

OpenEnv (Hugging Face). An open framework for creating reinforcement learning environments that are interoperable and shareable. Built as containerised services exposing a familiar agent interface, MCP and the Gym training protocol. It is designed to teach agents the capabilities they will use in the real world. For enterprises training custom agents, OpenEnv provides a standard way to build and share training environments rather than starting from scratch.

OpenSandbox (Alibaba). A general-purpose sandbox platform for AI applications with multi-language SDKs, unified sandbox APIs and Docker and Kubernetes runtimes. It includes a sandbox protocol defining lifecycle and execution APIs, so teams can extend it with custom runtimes. For enterprise security and operations teams, a standard sandbox format means agents can be tested in consistent environments before touching production systems.

Manik Surtani, CTO and co-founder of the AAIF, captured the significance: "Agent Skills, Open Responses, OpenEnv, and OpenSandbox each cover a different part of that stack, from how agents gain capabilities to where they run, and four different companies chose to contribute them to neutral governance. That is the job of this foundation: putting the community first by giving it one place to build shared standards, protocols, and infrastructure, with no single company controlling the outcome."

Why competing companies are choosing shared governance

The pattern is notable. Anthropic contributed MCP to the AAIF less than a year ago. Now it is contributing Agent Skills. OpenAI, which historically built proprietary APIs, is contributing an open specification for multi-provider model interfaces. Alibaba, one of China's largest technology companies, is contributing a sandbox platform. These are not peripheral projects. They are core infrastructure components.

The motivation is not altruism. It is market logic. Enterprise buyers are telling vendors the same thing: they will not bet on a single agent platform. They want agents from multiple providers to share context, call the same tools and run in governed environments. Vendors that participate in shared standards reduce integration friction for customers. Vendors that do not risk being isolated from a growing ecosystem.

David Soria Parra, co-creator of MCP at Anthropic, said: "Without the right context, even the most intelligent agents fall short. We launched Agent Skills a year ago to give developers and enterprises alike an open standard to make their agents more capable. Today, we are donating Agent Skills to the AAIF to ensure that anyone building with skills can shape what is next."

What this means for enterprise buyers

For procurement teams and IT leaders evaluating agent platforms, the AAIF expansion provides several practical signals.

Interoperability is becoming a procurement requirement. If agents from different vendors cannot share context, tools and execution environments, the enterprise ends up with a fragmented agent estate. The AAIF projects create common ground. Buyers should ask potential suppliers which AAIF standards they support and what their roadmap for adoption looks like. A vendor that cannot answer that question clearly is likely betting on a walled garden.

Vendor lock-in risk is being addressed at the infrastructure level. Open Responses, in particular, means that the interface between an application and a model provider is being standardised. If this standard gains traction, switching model providers becomes an operational change rather than a re-platforming exercise. For enterprises concerned about concentration risk with a single model provider, this is significant.

Agent testing and security are becoming standardised. OpenEnv and OpenSandbox together create a path toward consistent agent evaluation and secure execution. Today, most enterprises test agents in ad-hoc environments. Standard sandboxes and training frameworks mean security teams can apply consistent policies across agent platforms, and development teams can compare agent performance on shared benchmarks.

The standards race is accelerating. The AAIF has grown from three projects to 10 in a year. The speed of expansion suggests that the agentic AI industry is moving through the same standardisation phase that cloud computing went through a decade ago. Early adopters of these standards will have lower integration costs and faster deployment cycles than those who wait.

What this means for agent platform suppliers

For companies building and selling agentic AI platforms, the AAIF expansion sends a clear message: the market is building shared plumbing, and platforms that connect to it will have a distribution advantage.

Suppliers that adopt AAIF standards early can claim interoperability with a growing ecosystem of tools, models and environments. That is a competitive advantage in procurement cycles where buyers are specifically asking about standards support. Suppliers that ignore or compete against these standards risk becoming islands in an increasingly connected landscape.

The presence of Alibaba in the AAIF is also worth noting. It signals that the open agentic AI standards movement is not limited to Western technology companies. Enterprises with global operations or supply chains should expect their agent platforms to work across regions and regulatory environments. Standards that include contributors from multiple geographies are better positioned to support that requirement.

The Agentic Expo angle

Agentic Expo, at Olympia London on 23 to 24 March 2027, is where the platforms, standards and buyers of agentic AI converge. The AAIF expansion is exactly the kind of infrastructure shift that buyers need to understand before making platform decisions. By March 2027, when the exhibition opens, the standards landscape will have evolved further. The platforms on the show floor will be judged not just on what their agents can do, but on how well they connect to everything else.

For exhibitors, the message is clear: standards support is becoming a differentiator. For visitors, the implication is practical: when you walk the floor at Olympia, ask every platform vendor which AAIF standards they support. The answer will tell you more about their long-term architecture than any product demonstration.

Get Tickets Exhibit at Agentic Expo

Sources: Agentic AI Foundation / Linux Foundation, "Agentic AI Foundation Announces Four New Projects: Agent Skills, Open Responses, OpenEnv, and OpenSandbox", 22 October 2026; aaif.io.