The Linux Foundation has launched a new nonprofit open-source standards body for AI agents called the Agentic AI Foundation (AAIF), as it seeks to build shared protocols and frameworks around this emerging technology.
Membership in the AAIF has already tripled since it was unveiled in December, the foundation said, with AI companies contributing key foundational technologies such as Anthropic’s Model Context Protocol, Block’s Goose framework and OpenAI’s AGENTS.md.
Spearheading those efforts is David Nalley, director of developer experience at AWS who is the new governing board chair. With more than two decades in open source, including leadership roles at the Linux Foundation and the Apache Software Foundation, Nalley said the organization’s early focus will be on building the conditions for collaboration rather than dictating outcomes.
“When AAIF was being constructed, there was a ton of attention and interest. Everybody wanted to be a part of it,” Nalley said in an interview with The AI Innovator. “We’re very much looking for people who are willing to participate in thinking about the future of protocols and the future of tools and what is agentic AI going to look like in six months or five years.”
AAIF’s members typically fall into two groups: those from academia and elsewhere who are creating transformative technology and those who are using agentic AI to transform their organizations. “They want to be part of setting the direction and influencing the future,” Nalley said.
That effort comes at a moment of rapid change. AI is shifting from isolated models to agentic systems that can interact with software, data and other agents across enterprise environments. But each major vendor is building its own frameworks and protocols, raising the risk of silos. The foundation’s role is to provide a neutral venue where those underlying components can be developed collaboratively, allowing companies to compete higher up the stack rather than on basic infrastructure.
A different kind of standards body
“My top priorities really center around a couple of things,” Nalley said. “The first is, open source always has been a collaborative area where people get together and they debate about ideas and they implement them. And unlike a lot of standards organizations, you end up building a lot of reference implementations first and they become de facto standards.”
That philosophy reflects a deliberate departure from traditional standards bodies. Instead of formalizing specifications first, AAIF is encouraging developers and companies to build working systems and let standards emerge from what gains traction.
To support that approach, one of his immediate priorities is creating working groups across key domains such as security, privacy and emerging areas like agentic commerce. Those groups are meant to act as forums where competing companies and independent developers can align on shared problems.
“Standing up the working groups so that people can talk about security and privacy and agentic commerce and a number of other topics is really high on the list of things that are important,” Nalley said, noting that several groups have already begun work.
The rationale is rooted in a view that much of the foundational layer in agentic AI does not confer competitive advantage. Instead, common infrastructure that everyone agrees on can ensure development of interoperable systems. “That means we need a place to actively collaborate, have these discussions and have them in a vendor-neutral way,” he explained.
Carefully choosing members
Second, Nalley said AAIF wants to be deliberate in admitting members. AAIF has already seen membership tripling shortly after its launch, drawing organizations from across technology, finance and academia.
“We’re seeing a huge number of open source projects that want to join the foundation, and we want to be thoughtful” about admission, Nalley said. “One of the things I’m worried about is that people see membership in a foundation like AAIF as a Good Housekeeping seal of approval, … essentially an endorsement of both the technology and the project itself.”
As such, the foundation is formalizing criteria for project inclusion and maturity, including distinctions between early-stage and more established technologies. The goal is to give enterprises clearer signals about the most supported tools for production.
“It’s really important for the people who are going to consume these tools and protocols that we have a good way of saying, ‘this is a well-supported, really robust project and it has shown its maturity’ or ‘this is a really promising technology,’” he said.
Nalley also pointed to the risk of strategic drift as a key challenge. With broad interest from across the AI ecosystem, he said the foundation must stay focused on its core mission.
“I’m worried about misalignment on mission,” he said. “I’m worried that if we try and do everything, we will lose a lot of the impetus that’s driving us right now.”
But he pointed out that a shared mission doesn’t mean they will all agree on tactical implementations.
“It would probably be a bad idea if we were all operating in a hive mind and all nodding our heads at the same time,” Nalley said. “Some dissent and differing opinions are really important.”
The ‘lazy consensus’
Managing collaboration among direct competitors is another defining feature of AAIF. Members include companies that compete aggressively in AI and cloud markets, for example, and yet are working together on shared infrastructure.
Nalley said this dynamic is not new in open source and has historically accelerated innovation. He pointed to foundational internet technologies such as HTTP as an example. “If everybody had created their own version, we would be in a much worse space,” he said.
He also cited tools like PyTorch, widely used across competing AI companies, as evidence that shared platforms can reduce duplication and speed development.
“The fact that they’re all using it essentially means they didn’t have to go recreate it themselves − saved themselves a bunch of time,” he said.
The foundation’s governance model is designed to move quickly. Nalley said most technical decisions are made through “lazy consensus,” where contributors propose changes and proceed unless objections arise.
“The overwhelming number of technical issues are typically decided via lazy consensus,” he said.
The approach prioritizes speed over formal process, reflecting the rapid pace of AI development. Traditional standards procedures could take a year to 18 months – at a fast track. The pace of AI doesn’t allow AAIF to delay standards for that long.
“The world would have changed considerably by the time you got a standard ratified in the traditional manner,” Nalley said.
He acknowledged that the result can be uneven, with standards evolving through multiple versions over time. Nalley argued that versioning and iteration allow the ecosystem to adapt without stalling progress.
‘Growing faster than anything else’
Looking ahead, he said success for the foundation will be measured by its impact over decades. Longevity, participation and real-world impact are the key indicators he will watch.
“First is continued existence,” he said. “I’m not talking about two years or five years, but 10, 20 years down the road. If the foundation still exists and still has lots of members and lots of activity, that’s a good sign.”
Ultimately, however, the test will be whether the foundation’s work is widely adopted. “The real question is do the projects that are housed at the foundation matter?” Nalley said.
For now, the early surge of interest suggests the industry sees a need for coordination as AI systems become more autonomous and interconnected. Nalley said the pace of growth has been striking even by open-source standards.
“I’ve been in and around foundations for a large part of my career, and I’ve not seen anything that has this type of growth,” Nalley said. “It’s growing faster than anything else I’ve seen.”
Be First to Comment