Agent-led adoption is the next domain that needs this — and the window to define it is open right now.
Not a runtime. Not a marketplace. A document — and the schema that lets every system in the chain read, write, and trust it.
The same document threads through all six steps — no re-keying, no per-vendor connectors, no lost context.
If their agents are autonomously adopting tools, they need a single audit trail that survives every vendor change.
Speaking OCP turns a months-long enterprise eval into a packet the buyer's policy engine can read in seconds.
OCP is the lingua franca that lets capabilities flow between platforms without re-certifying each one.
The narrowest, sharpest form of an OCP document is the artifact an agent produces when it tests a new capability — the bench results, costs, scopes, and signatures.
Enterprises will pay for it before any standard is ratified, because their agents are already running these evaluations and producing nothing portable.
Every packet our customers ship into their audit log is a working draft of the standard.
The first authority that issues the canonical validator becomes the reference implementation everyone else integrates against.
Once a handful of large enterprises require an OCP packet from every capability under evaluation, the long tail of vendors follows in quarters, not years.
The packets our customers produce form the largest corpus of grounded agent-evaluation evidence in existence — defensible, anonymized, and licensable.
By spinning the spec into a foundation early, we become the convener — the layer no replacement can route around without breaking the world.
Tool calls are no longer human-supervised. The first generation of fully agent-led adoptions is in production at top-50 banks right now.
The EU AI Act, NIS2, and forthcoming US executive guidance all imply an audit trail that doesn't yet exist in standard form.
Procurement vendors, agent platforms, and MCP authors each see one face of the problem. The neutral seat is empty for ~18 months.