Why do some platforms wall everyone in, while others open up and let outsiders plug in freely?

This explores why platforms sometimes lock their ecosystems down and other times open them to shared standards. The short version from the corpus: closed systems are where a platform captures value, and open standards are where everyone else does. The interesting part is that the pressure toward openness usually comes from outside the platform itself, from rivals, regulators, or the plain cost of getting anyone to adopt something new.

The pull toward closure is about extracting surplus. Doctorow's enshittification lifecycle describes the pattern. A platform first gives users a good deal to attract them, then squeezes the businesses that depend on it, then squeezes everyone for shareholders Do platforms inevitably decline through value extraction cycles?. A formal economic model shows the same pattern on a small scale. An AI platform that charges for 'inspection tokens' (the right to look at sellers) has a profit incentive to price them so buyers over-search. Buyers end up no worse off, but money moves from sellers to the platform and the market as a whole gets less efficient Does platform pricing of inspection tokens encourage excess search?. That kind of extraction is only possible when the platform controls the gate. Waller expects agentic commerce to split the way card networks did: some platforms will let any agent in, while others will require their own agent. That choice will determine how concentrated the market becomes Will agentic commerce platforms adopt open or closed models?.

The push toward openness tends to come from those who would lose in a winner-take-all race. O'Reilly argues that protocols are never neutral plumbing. They shape markets. He gives MCP as an example: a shared protocol turned a race to own the chatbot platform into a web of vendors who can all compete Do protocols reshape markets by enabling multi-vendor competition?. Regulators reason the same way. The UK's Competition and Markets Authority treats data portability as basic market infrastructure, not a convenience feature. Its point is that if your data can't move with you, an AI agent can't act for you in any meaningful way, and switching providers stops being a real option Should data portability be infrastructure for agent-driven markets?.

Two results complicate the 'open good, closed bad' story. First, the open standards that win usually don't try to replace what's already there. Coordination protocols spread by wrapping and bridging existing ones like MCP, so that value builds up gradually and nobody has to rewrite everything at once Should coordination protocols wrap existing systems or replace them?. Second, openness has real engineering costs. In production workflows, tool access through MCP caused unpredictable failures: the agent picked the wrong tool or guessed parameters badly. Teams fixed this by switching to direct, hard-coded function calls, and a survey found 85% of production teams build their own custom agents rather than use shared frameworks Why do protocol-based tool integrations fail in production workflows?. Some 'closedness' comes from wanting things to be reliable, not from wanting to extract value.

Cost works the same way. A cost model of self-hosting LLMs shows small models paying for themselves within months When does self hosting an LLM beat commercial API costs?. That suggests customers can keep some leverage against closed API providers by running models themselves. The corpus is thinner on why a dominant platform would choose to open up on its own. Most of the material casts openness as something competitors and regulators push for, not something incumbents choose.