Your agents source from outside suppliers. You approve what binds.

What every new supplier costs today

If your procurement agents already deal with outside suppliers, you know the pattern. Each new supplier relationship is wired by hand — message formats, negotiation logic and connection code built and tested from scratch, then owned and maintained afterwards. Independent vendor estimates converge on roughly $50,000 to $150,000 per year for a single custom enterprise integration, maintenance included. That expense recurs for every supplier and never fully amortises.

The arithmetic scales badly. A programme covering ten or more systems is estimated at $400,000 to $1,200,000 to build, with a further $120,000 to $350,000 a year to maintain. Taking the lower per-connector figures on their own, a company transacting with twenty counterparties faces on the order of $1.5m in the first year, and one at fifty counterparties around $3.75m — the build cost plus a year of maintenance, multiplied by the number of relationships. The line item grows with the count of suppliers, not with the value they bring.

Subchain removes the multiplier. Your agents propose, counter and accept machine-readable terms against one standard, so a supplier is configured once rather than integrated repeatedly. The integration you own is to the protocol, not to each counterparty; adding the next supplier is a configuration step, not another bespoke build.

There is a second cost that rarely appears on the invoice. As agent deployment accelerates, most organisations cannot say with confidence what their agents are authorised to commit to. One practitioner survey of senior technical leaders found that 85% of organisations have no formal accountability for AI agent behaviour, and that no single pre-deployment control — a named accountable person, a security review, a pause or revoke process, scoped access — is used by even 40% of them. For a procurement function answerable for supplier terms, that gap is the real exposure, and it is the gap the approval gate is built to close.

The boundary, and the honest open question

For a procurement function, the boundary matters before the mechanism. No agent closes a supplier deal end to end on its own: the runtime settles terms, but a named person clears the gate before anything binds. It is not a signing autopilot, and the immutable log it keeps is a record of authorisation — not a distributed chain, a token or a consensus network. It governs the residual risks around agent authority, approval-gating and liability that current deployments leave unresolved.

On enforceability, one primitive is already settled. Automated contract formation has had legal footing since 2005 under UNCITRAL Article 12, which materially de-risks the question of whether a machine-negotiated agreement can be valid at all. What remains unsettled in many jurisdictions is how liability is allocated when an agent errs. Where an agreement's authority-to-bind may not hold, the runtime is designed to escalate it for human legal determination rather than let it stand.

We are equally direct about what we have not yet proven. Demand for a governed-negotiation layer is currently inferred from the integration pain around it, not yet confirmed by signed design partners or buyer conversations. Turning that inference into evidence — by signing the first procurement teams as design partners — is the work of this stage. We would rather state that plainly than imply a scale we do not have.