Standards outlive platforms

The primary source on the standards moment of the year: Anthropic handed the Model Context Protocol to the Linux Foundation’s Agentic AI Foundation, co-founded with Block and OpenAI, with AWS, Google, Microsoft, Cloudflare and Bloomberg in the platinum tier. The numbers explain why: over 10,000 active public servers and more than 97 million SDK downloads a month. My uncomfortable reading: this was not given away out of altruism but at exactly the right moment — once the protocol had already won and neutral governance was worth more than control. That is smart strategy, and it deserves to be called that.

The industrial answer that almost nobody in the AI feed picks up: the OPC Foundation is making over 430 Companion Specifications RAG- and MCP-ready, positioning OPC UA as the semantic layer between agents and the plant. Exactly the gap we discussed here in July — an agent is only as good as the context it can reach. The difference from the software world: these information models have existed for years, from CNC machines to packaging, with cross-vendor semantics. Industry does not have to invent the context. It only has to make it reachable.

How an alliance does standardisation properly, with dates rather than declarations of intent: satellite discovery and walk-by reading in 2026, crypto agility in 2027, a standard application data format in 2028. The last one is the most interesting — it removes exactly the custom integration that makes every LoRaWAN project more expensive than it needs to be. I will be looking at this closely in Lyon and Amsterdam in September. Over ten years of alliance discipline, incidentally, is why LoRaWAN is still here while much from the same era is not.

The most uncomfortable read of the week, and the most historically literate: Wilson traces railroad gauges through "embrace, extend, extinguish" to Apple’s handling of Android messaging, showing how reliably standards get captured. His line belongs in every architecture review: every choice between an open standard and a proprietary API is a bet on the future behaviour of the platform that owns the API — and history says it is a bad bet. I spent four years on the OSGi board and four on the Eclipse steering committee. The technically better standard does not win. The one with better distribution wins.

And the fact-based critique to close, the one missing from the protocol euphoria: the authors test five interoperability protocols against six governance dimensions and find that voting and the preservation of dissent are absent in all of them, with deliberation partial at best. Their conclusion: governance for agent communities is a missing architectural layer above today’s standards — not something you retrofit with a few fields in a spec. Anyone wiring agents together today is wiring technology without a constitution. In regulated industry that is precisely the question procurement will ask.