Truly Open
As always, the comments on a Slashdot post are the whole reason why I enjoy reading the content. The coverage of the new Agentic AI Foundation (AAIF) co‑founded by OpenAI, Anthropic, and Block under the Linux Foundation, triggered exactly the debate you’d expect from a community that has spent decades defending real openness. While the headline announcement talks about neutral governance and open standards for agentic AI, many commenters zeroed in on the gap between the branding and the reality of what’s being “opened.”
The core skepticism is simple: standards without real openness around models and data can look like window dressing. As NERDS.xyz’s Brian Fagioli argues (and Slashdot highlights), the companies are donating artifacts like AGENTS.md, MCP, and goose. All useful, but essentially “safe” components that don’t threaten their core proprietary advantage. In the comment threads, users echo this, comparing these contributions to getting a free MP4 player while the actual film library stays locked behind a rental wall: you can wire agents together, but you can’t meaningfully re‑create or audit the systems that matter.
Several Slashdotters draw a sharp line between open standards and open AI. A “real open project,” one commenter notes, would ship not just protocols and SDKs but the full training recipes: datasets in usable form, training code, and reproducible specs, so an independent team could, at least in principle, rebuild the system from scratch. By that yardstick, OpenAI and Anthropic are still firmly in the “closed AI” camp, no matter how interoperable AGENTS.md or MCP become. Another commenter points out that there are a few genuinely open efforts (they mention projects like Olmo) that do share full training pipelines, underscoring that the bar for openness is higher than what AAIF’s founders are currently clearing.
Others zoom in on the Linux Foundation angle. One top‑rated comment argues that it makes sense for the Foundation to engage as AI “is not going away,” and Linux users will want sane, standard ways to plug it into their systems. The “WTF,” in this view, is not the Linux Foundation agreeing to host AAIF, but what OpenAI and peers will actually do inside the new body: will they truly cede control to a community process, or simply use the Foundation’s neutral brand as cover while steering standards toward their own stacks? The pattern here of big vendors using “open” language to shape ecosystems before grassroots projects can, is one Slashdot readers feel they’ve seen many times before. They know how this story ends.
Notwithstanding the skepticism, interoperability is the one area where commenters concede there’s real upside. The promise that agents built with goose can talk to MCP servers and follow AGENTS.md across tools and clouds is genuinely attractive for developers trying to avoid a “one‑vendor agent stack.” But even here, the mood is wary: people like the idea of fewer incompatible protocols, yet worry that if the initial standards are effectively defined by a small set of closed‑model providers, the ecosystem will bake in subtle forms of lock‑in anyway.
While this may not be a blanket rejection of AAIF but it is an early warning system. The community seems to be saying: standards for agents are necessary, the Linux Foundation is a sensible home, but calling this “open” AI is premature as long as the most powerful models and training data remain black boxes. Until that changes, many free‑software veterans will see moves like AAIF as “smoke and mirrors” just a way for Big AI to write the rules of the next era while giving away just enough to claim the mantle of openness.