Follow ZDNET: Add us as a preferred source on Google.
ZDNET key takeaways
- Major Agentic AI players are backing AAIF.
- Open source standards enable cross-LLM agents.
- AAIF currently uses MCP, Goose, and AGENTS, with more to come.
At Open Source Summit Japan in Tokyo, the Linux Foundation announced the launch of the Agentic AI Foundation (AAIF). This new, open standards industry-backed consortium seems destined to be a game changer.
Open, interoperable, and governed
The AAIF’s mission is to standardize and accelerate the emerging ecosystem of AI agents. Backed by OpenAI, Anthropic, Block, and a who’s who of major cloud and software vendors, the group aims to make “agentic AI” infrastructure open, interoperable, and governed by familiar open-source norms.
Also: Why AI agents failed to take over in 2025 – it’s ‘a story as old as time,’ says Deloitte
Announced under the Linux Foundation’s umbrella, the AAIF will be a neutral home for open-source projects focused on AI agents.
Agents are systems that can plan, act, and coordinate tools or other agents between Large Language Modules (LLMs) and other information services on your behalf. The foundation’s mandate is to ensure that this new layer of AI infrastructure evolves with transparent governance, cross-vendor standards, and community-driven development, rather than proprietary silos.
To make this happen, the AAIF is built on three cornerstone technologies: Anthropic’s Model Context Protocol (MCP), Block’s Goose Coding Agent, and OpenAI’s AGENTS.md specification for describing and orchestrating agents. All these projects have now been donated to the foundation.
Also: 3 ways AI agents will make your job unrecognizable in the next few years
Together, they’re meant to make a new shared software stack. In this stack, MCP will act as a universal protocol for connecting models to tools and data. AGENTS.md serves as an open format for defining agent capabilities, and Goose is a real-world agent implementation built on these concepts.
A universal way of dealing with agents
This is a big deal because agentic AI, which consists of multi-step, tool-using agents rather than simple chatbots, despite all the agentic hype, has big problems in the real world.
As the data protection company Rubrik‘s chief product officer, Anneka Gupta, recently pointed out, agentic AI can make horrible mistakes. Just as bad, if not more so, because agents can act as users, they can cause havoc.
Also: AI agents are already causing disasters – and this hidden threat could derail your safe rollout
Say, for example, someone in accounts payable uses an agent to pay freelancers. That’s a perfectly reasonable use. But what if someone else can use that same agent? The system may not know that it’s Joe, the janitor, rather than Angela, the CFO, asking for a million dollars to be paid to George, the janitor’s husband. Sounds silly? This kind of thing will happen.
We need a universal way of dealing with agents, and AAIF will be part of the solution. AI agents are moving very quickly from labs into production. Companies are pitching agents to automate workflows, software development, and data-heavy operations.
Without standard protocols and formats, each vendor’s agent stack risks becoming a separate island. This makes it harder for enterprises to mix tools or switch providers. It also means securing multi-service agents will be harder than ever.
Also: Should you trust AI agents with your holiday shopping? Here’s what experts want you to know
As Chris DiBona, vice president of Microsoft’s office of the CTO, said in a statement, “For the agentic future to become a reality, we have to build it together, and we have to build it in the open. The AAIF will give industry and developers a shared and transparent path to evolve the agentic AI ecosystem. Microsoft remains committed to supporting this journey to create an open, interoperable, and reliable foundation for everyone building and using agents.”
In the near term, the Agentic AI Foundation will focus on evolving MCP, AGENTS.md, and Goose under open governance, while recruiting additional projects that fit into a shared agentic stack.
In the long term, members expect AAIF to become the central venue for interoperability profiles, security and evaluation frameworks, and reference implementations that vendors and open-source projects can build on as agentic AI becomes part of mainstream infrastructure.
