The settlement rail for autonomous agents is solved. The attribution layer underneath it is not — and it is the layer that decides who gets paid.
Execution is no longer the constraint. x402, ACP, MCP and AP2 have made machine-to-machine payment a solved problem, and stablecoins already move more value than the card networks. The question has moved downstream.
Agents cannot dynamically split fees on demand. Multi-party settlement currently relies on hardcoded partner splits — decided by a human, in advance, before anyone knows which agents will actually contribute to an outcome.
That is fine when two parties transact once. It fails the moment a purchase is the product of six agents, three models and a retrieval step — because the split has to be computed after the work, from what each participant actually contributed.
Every protocol in the agentic stack answers "did value move?" None answers "whose cognition caused it?" Volume grows; the attribution question gets harder in proportion.
A static split is a guess made before the work. At 25% of global transactions, guessing is not a rounding error — it is the difference between a functioning agent economy and one where only the platform gets paid.
The industry built rails first. What is absent is a meter — a way to record contribution per inference, per agent, per outcome, that settlement can then read from.
METAVENTIONS AI builds the cognitive metering layer: UCW captures contribution at the point of inference, Proof of Cognition makes that record verifiable, and settlement reads the split rather than assuming it. Cognitive equity — value accruing to whoever did the thinking, not only to whoever owns the endpoint.
"Your rail settles the transaction. The open question is who the transaction should have paid. That's the layer we build."
This brief is the wedge. The primitive underneath it — cognitive extraction, the UCW architecture, Proof of Cognition, $META and BASIX — is the whitepaper. METAVENTIONS Whitepaper v1.0 (PDF).
The dynamic-split limitation is documented by AWS. A metering layer resolves it without changing the rail.
Execution is not the bottleneck at $3.5T/30d. Attribution is.
Mosaic solves identity for agents. Metering is the monetary complement, not a competitor.
A decentralised agent network needs an answer to intra-network value distribution.