Back to Blog
Open StandardsAI AgentsMCPSoutheast Asia

The Agentic AI Foundation: Why Open Standards Matter for Southeast Asian Developers

June 13, 2026 by Inference Loops

When two fierce competitors agree on something, pay attention. In June 2026, OpenAI and Anthropic — companies that spend their days trying to out-ship each other — jointly co-founded the Agentic AI Foundation (AAIF) under the neutral umbrella of the Linux Foundation. The goal: shared, open standards for how AI agents work, talk to tools, and interoperate.

This is the kind of news that scrolls past most developers as governance noise. It shouldn’t. The arrival of credible, vendor-neutral standards is one of the most consequential things that can happen to a young technology — and for developers in Southeast Asia specifically, it tilts the playing field in a direction the region has rarely enjoyed. Let’s unpack what happened and why it matters here.

What just happened

The Agentic AI Foundation is a neutral home, hosted by the Linux Foundation, for the standards that let agents from different vendors work together. The Linux Foundation is the same institution that stewards Linux itself, Kubernetes, and a long list of the infrastructure the modern internet runs on. Putting agent standards there is a deliberate signal: these protocols should belong to the commons, not to any single company’s roadmap.

The significance is in who co-founded it. OpenAI and Anthropic are direct rivals at the frontier. For them to jointly hand agent standards to a neutral foundation is an admission that interoperability is more valuable to everyone — including them — than lock-in. It’s the agent era’s version of the moment the web settled on open HTTP and HTML instead of a dozen incompatible corporate networks. Standards are how a technology stops being a collection of walled gardens and starts being an ecosystem.

The standards already winning

The Foundation didn’t appear in a vacuum. It’s institutionalizing conventions that developers had already adopted from the bottom up — which is the healthiest way for a standard to arrive.

AGENTS.md — the README for agents. This is a simple convention: a plain markdown file in your repository that tells coding agents what they need to know to work in it — how to build, how to test, what conventions to follow, what not to touch. Where README.md is written for humans, AGENTS.md is written for agents. It has spread to over 60,000 repositories, and the reason is obvious to anyone who has watched an agent flail in an unfamiliar codebase: a few lines of orientation dramatically improve what the agent produces. It costs nothing and works with any agent that reads it — which is the whole point of a standard.

MCP — the USB-C for AI tools. The Model Context Protocol is an open standard for connecting agents to tools, data sources, and services. Instead of every tool building a bespoke integration for every agent, a tool exposes one MCP server and any MCP-aware agent can use it. The ecosystem has grown to roughly 9,400 servers — connectors for databases, APIs, file systems, SaaS products, internal services. That number is the real story: it’s a network effect compounding. Each new MCP server makes every MCP-aware agent more capable, and each new agent makes building an MCP server more worthwhile.

Together, AGENTS.md and MCP are doing for agents what HTTP and USB did for their domains: turning N×M custom integrations into N+M standard ones.

Why open standards matter

Step back and the general case for open standards is about four things, all of which compound:

  • No vendor lock-in. If your agents, tools, and instructions speak open protocols, you can swap the model or the harness underneath without rewriting everything. Your investment survives the next vendor shake-up — and in a field where the leader changes every few months, that’s not a small thing.
  • Interoperability and network effects. Standards let independently built pieces snap together. The value isn’t in any one MCP server; it’s in the 9,400 of them working with any compliant agent. The whole becomes worth far more than the sum.
  • Portability of skill. When the conventions are open and shared, what you learn building against them transfers everywhere. Your MCP and AGENTS.md fluency isn’t tied to one employer’s stack.
  • Lower barriers to entry. You don’t need a partnership, an enterprise contract, or special access to participate. You read the spec and build. That’s the most democratizing property a technology can have.

That last point is where this stops being abstract and starts being regional.

Why this matters more for Southeast Asia

Open standards are good for everyone, but they are disproportionately good for developers and companies outside the incumbent power centers — and that’s exactly where Southeast Asia sits.

Consider what lock-in costs a smaller market. When a technology is controlled by a few vendors and accessed through gated partnerships, the advantage flows to whoever has the relationships, the enterprise budgets, and the proximity to the vendor. A team in Phnom Penh or Jakarta starts several steps behind a team in San Francisco — not on talent, but on access. Open standards delete that gap. The MCP spec is the same spec everywhere. AGENTS.md works identically in every repository on earth. There is no regional tier, no gatekeeper, no “enterprise edition” you need a Bay Area network to unlock.

Make it concrete. Two years ago, integrating an agent with a company’s internal systems meant a custom build against whatever proprietary SDK the vendor offered — work that favored teams with vendor relationships and big budgets. Today that same integration is an MCP server: one open spec, readable by anyone, usable by every compliant agent. A two-person shop in Da Nang can wire an agent into a client’s stack as cleanly as a hundred-person consultancy in London, because they’re both reading the same public protocol. The barrier that used to be “who do you know and what can you afford” is collapsing into “did you read the spec.” For a region long stuck on the wrong side of that barrier, that’s transformative.

This compounds with the argument we’ve made throughout this blog. In the three-layer model — harness, loop, orchestration — MCP and AGENTS.md are plumbing at the harness layer, and they’re now open plumbing. That means a regional developer can adopt a commoditized, standards-based harness and pour their energy into the layer that actually differentiates: engineering good loops. The expensive, lock-in-prone parts are becoming free and universal; the part that rewards judgment is the part that’s left. That is the leapfrog, handed to the region on a plate.

There’s also a contribution angle worth naming. Open standards aren’t just something you consume — they’re something you can shape. An MCP server for a Khmer-language service, a regional payments API, a local government data source — built by a Southeast Asian developer — becomes instantly usable by every agent in the world. The region doesn’t have to wait for a foreign vendor to support its tools and languages. Under open standards, it can build that support itself, once, and have it work everywhere.

The honest caveat

Standards are not a guaranteed happy ending. They can be captured by the largest players, fragmented by competing factions, or weighed down by design-by-committee. The fact that OpenAI and Anthropic co-founded the Foundation is encouraging, but the same incumbents have the most leverage to steer it. The healthy signal is the bottom-up adoption — 60,000 repos and 9,400 servers are developers voting with their code, and that’s much harder to capture than a committee. The trajectory is good. It still has to be defended, and the way you defend it is by using the open standards instead of the proprietary alternatives.

What this means for you

If you’re a developer or team in the region, the practical takeaways are concrete:

  1. Adopt the open conventions now. Put an AGENTS.md in your repos. Build and consume MCP servers. Fluency in these is becoming table stakes, and it’s fully portable.
  2. Build on the standard layer, differentiate above it. Treat the harness and its open plumbing as commodity. Spend your effort on the loop and the judgment — the parts that don’t commoditize.
  3. Contribute, don’t just consume. The tools, languages, and data sources specific to this region are an opening. Build the MCP servers no foreign vendor will prioritize.

Conclusion

The Agentic AI Foundation is a quiet milestone with loud implications: the agent era is choosing open standards over walled gardens, and rival giants are funding the commons because interoperability beats lock-in even for them. For Southeast Asia, that’s rare good news — it erases the access gap that has historically held regional developers back, leaving talent and judgment as what matters. The plumbing is becoming free and universal. The opportunity is to build on top of it.

That’s the bet we make at Inference Loops: open standards at the base, world-class loop engineering on top, built from Southeast Asia. If your team wants to build on the open agentic stack — and get the layer above it right — let’s talk.