You’re building an AI agent. You’ve heard the news: OpenAI and four of its biggest rivals just agreed on a common standard for how agents communicate. The headlines call it a victory for interoperability. A win for the ecosystem. A step toward a unified future.
Stop right there. That’s the story they want you to believe. The real story is far more unsettling.
This isn’t about making your life easier. It’s about who gets to own the air traffic control for every AI agent that will ever fly. And the answer might be the same five companies that just shook hands.
Open standards aren’t always open. Sometimes they’re the most elegant moats ever built.
Let’s rewind. The standard in question is called the Model Context Protocol (MCP). It defines how agents discover and invoke skills, how they pass data, how they authenticate. On paper, it’s beautiful: no more walled gardens, no more rewriting integrations for every platform. Newcomers can plug in overnight. Fragmentation disappears.
But here’s the twist that almost nobody is talking about: the entity that controls the standard controls the gateway. Every agent that wants to play in this ecosystem must pass through the protocol’s choke points. Who decides what skills are allowed? Who sets the rules for authentication? Who updates the spec when a new threat emerges?
In the best case, a benevolent steward. In the worst case, a gatekeeper that can slowly tighten the screws.
You’ve seen this movie before. Remember Manifest V3? Google’s ‘open’ extension standard that quietly killed ad blockers? Remember how ‘open’ HTML5 eventually gave the browser vendors the power to dictate what apps could do?
When your rivals agree to your standard, you haven’t made peace. You’ve made them tenants on your land.
OpenAI didn’t propose MCP because they’re altruistic. They proposed it because they saw the writing on the wall: every startup was building its own agent protocol, and the market was heading toward chaos. Chaos favors no one. A single standard, controlled by the incumbents, commoditizes the periphery while locking in the core.
Consider the math. OpenAI brings the largest user base, the most developer mindshare, and the deepest pockets. Anthropic brings safety credibility. Google brings infrastructure. Microsoft brings enterprise distribution. When these five agree on a standard, it becomes the de facto norm. Any startup that tries to build a competing protocol is immediately labeled ‘incompatible.’ Any developer who wants to reach the biggest audience must comply.
That’s not a standard. That’s a toll booth.
I spoke with a developer who builds autonomous agents for logistics. He said, ‘I thought it was great until I realized I’d need their permission to run my agent on their infrastructure. Suddenly, I’m not building a product. I’m building a plugin.’
This is the anxiety that the press releases gloss over: the relief of a unified ecosystem comes with a silent cost — your autonomy. You’ll be free to build, as long as what you build fits within the protocol’s bounds. And those bounds will be set by companies whose primary interest is protecting their own moats.
Neutrality is death. Pick a side. Either this standard is the best thing to happen to AI agents, or it’s the first step toward a new kind of feudalism. I’m betting on the latter.
Does that mean you should boycott MCP? Not necessarily. But you should treat it with the same skepticism you’d give any proprietary standard dressed in open clothing. Ask yourself: who writes the rules? Who enforces them? Who profits when the network grows?
The most dangerous technology isn’t the one that’s closed. It’s the one that’s just open enough to make you feel free, while quietly building a cage.
The question isn’t whether you’ll adopt this standard. It’s whether you’ll realize you’re building on someone else’s land — and they’re already drafting the lease agreement.
FAQ
Q: Isn't this just a standard like any other? Why is it dangerous?
A: Most standards are genuinely open. But when a small group of dominant players controls the protocol, they can subtly change the rules over time—prioritizing their own services, throttling competitors, or requiring approval for new capabilities. History shows that 'open' standards often become tools for gatekeeping.
Q: What should a developer do right now?
A: Adopt the standard for compatibility, but never build your entire business on it. Always maintain a fallback path. Keep your core logic decoupled from the protocol. And watch the governance: if the standard is controlled by a single entity or a small club, start planning for an alternative.
Q: If it's such a power grab, why would companies like Anthropic and Google agree to it?
A: Because they each believe they can influence the standard more than their rivals. They'd rather be at the table than on the menu. Plus, the alternative—thousands of incompatible protocols—was worse for everyone. They're playing a long game, betting that control of the protocol gives them leverage over the next wave of AI.