Do protocols reshape markets by enabling multi-vendor competition?

O'Reilly, writing with Ilan Strauss and Sruly Rosenblat as part of their AI Disclosures Project, argues that protocols are not neutral technical plumbing but mechanisms that shape who can compete in a market. He traces the claim through the early AI chatbot market, which he describes as "a winner-takes-all race to be the dominant platform for AI in the way Windows became the platform for PCs, or iOS and Android for phones. Whoever wins controls the market." He then points to Anthropic's introduction of the Model Context Protocol (MCP) as the hinge: "All of a sudden, the landscape looked more like a web. There could be many winners. It didn't matter what model you were running or whose APIs you were calling as long as you followed the protocol." The general claim he draws from this single case: "Protocols are not just technical infrastructure. They are market-shaping mechanisms."

The reasoning runs through two distinctions borrowed from Summer of Protocols writer Venkatesh Rao and Ethereum researcher Danny Ryan. First, a protocol is a "stratum of codified behavior" that becomes invisible once adopted — a "Whitehead advance" after philosopher Alfred North Whitehead's point that civilization advances by extending what people can do without thinking. Second, Rao contrasts a protocol, an "engineered argument," with an API, an "engineered agreement" enforced by one dominant actor — the difference being who holds power over the terms of interaction. O'Reilly folds in a further distinction between de jure protocols that arise from a formal standards process and de facto protocols that arise through market contention; MCP falls in the latter camp, which is part of why he treats disclosures themselves as functional protocols — observability points, like credit-card fraud monitoring or email spam filtering, that make a market's behavior inspectable and therefore governable.

This sits one level above the protocol notes already in the library, which work the formal-design register: Can classical agent architectures fix agentic AI accountability? argues agentic AI needs BDI architectures and formal commitment semantics to become verifiable, and Can agent protocols be efficient, versatile, and portable simultaneously? classifies existing protocols by the versatility-efficiency-portability trade-offs their schemas encode. Both ask what makes a protocol technically sound. O'Reilly's claim is adjacent but different in kind: regardless of its technical design, a protocol's mere existence and adoption can convert a winner-take-all platform race into a multi-vendor web, which is a competition-policy effect, not an engineering one.

The argument is a single-case narrative from a Substack essay, not a peer-reviewed or empirical study, and O'Reilly does not measure market concentration before and after MCP's introduction or rule out other causes of the AI ecosystem's multi-vendor shape (new entrants, falling model costs, open-weight releases). The piece also does not say whether de facto protocol adoption reliably prevents re-concentration — a protocol-following "web" can still produce a dominant hub. The implication it supports is narrower than "protocols democratize markets": that protocol choice is a lever competition-policy thinking should track, worth watching rather than treating as settled.