The Protocol War Is a Switchboard War
On July 11, The Information reported that five of the largest companies in enterprise software had quietly agreed to back the same thing.
Google. Microsoft. Salesforce. Snowflake. ServiceNow. The two clouds most enterprise data runs on, plus the systems that hold the customers, the data, and the workflows. All lined up behind a shared protocol for how AI agents connect to enterprise systems — and pointed, explicitly, at Anthropic and OpenAI.
The industry read it the same way within a day: this is the protocol war. Anthropic’s Model Context Protocol became the default for tool connections over the last eighteen months, and this is the incumbents’ answer. Whoever owns the standard that decides how agents reach data, tools, and each other gets default status in every enterprise deployment for the next decade. TCP/IP. HTTP. The plumbing decides who owns the era.
The framing is right about the stakes. It’s fighting over the wrong layer.
What they’re actually fighting over
Strip the press releases and every contender answers the same class of question.
MCP: how does an agent reach a tool. A2A: how does one agent reach another. Cloudflare’s x402: how does an agent pay a paywall it hits. Different verbs, one category. Reach. Exchange. Transact.
That’s connection. Connection is necessary — nothing works without it. But it was never the thing that’s missing. A connection standard tells an agent how to get into your CRM. It says nothing about what the agent committed to once it was in there, whether that commitment was fulfilled, or who is accountable if it wasn’t.
Read that again. You can wire the entire enterprise together with a single, beautifully coherent protocol and still not be able to answer: across all of these agents, what did they commit to on my behalf, and was it done.
Routing a message to the right system is not the same as coordinating what’s owed across all of them. One is a switchboard. The other is infrastructure.
A better switchboard makes it worse
Here’s the part the protocol war gets backwards.
OutSystems surveyed 1,900 IT leaders this month: 96% already run AI agents, 94% now call agent sprawl a security risk, and 12% have any centralized way to manage it. IBM projects the average large enterprise will be running around 1,600 agents by year end and that most can’t govern the ones they already have.
Now connect all of them with one clean, universal protocol.
Perfect connection across 1,600 ungoverned agents doesn’t reduce the failure rate. It removes the friction that was slowing the failure down. A frictionless switchboard for untracked commitments is just faster untracked failure — at machine speed, across every system at once.
The protocol war is optimizing the one layer that was never the bottleneck.
The regulator that might have built the other layer just stepped back
For a while there was an external forcing function.
The EU AI Act’s high-risk obligations — record-keeping, traceability, human oversight, documented accounts of what an automated system did and on whose authority — were set to bite on August 2, 2026. That’s the accountability layer, mandated into existence.
On June 29, the Council gave its final sign-off to the Digital Omnibus. Those obligations move to December 2, 2027. Sixteen months.
So the market is racing to standardize the connection layer, and the one regulator that might have mandated the accountability layer just pushed it past the horizon. Deployment is waiting for neither. 96% is already in.
Nobody in this story is building the layer that holds what your agents owe.
The wire is being standardized by five companies with more resources than most governments. It will get built with or without us. It doesn’t need us.
The layer above the wire is a different thing. The one that holds what every agent committed to — across every system they’re now perfectly connected to — and tells you whether it was closed. That isn’t a protocol you adopt or a deadline you comply with. It’s infrastructure someone has to build.
It’s still nobody’s product.
That’s what we’re building.
