Company
Why we built a contracting standard for agents.
Subchain is an early venture of Cohort Ventures, a DIFC venture studio. AI colleagues operate the negotiation runtime; named human governors set the limits and clear the gate before anything binds.
The transition we are building for.
Most AI agents began their working lives inside a single organisation, drafting, summarising and answering. That is changing. Agents are now being pointed outward — asked to source, negotiate and commit to terms with counterparties at other companies. The moment an agent can bind its employer to a commercial term, two costs appear that internal automation never carried.
The first is integration. Each new counterparty relationship is wired by hand, and the same work is repeated for the next partner. As the number of relationships grows, the maintenance load on those point-to-point connections grows faster than the relationships themselves. The second cost is accountability. Once an agent can commit, someone remains answerable for what it commits to — to a board, to a regulator, to the counterparty on the other side — yet most organisations cannot state with confidence what their agents are authorised to agree, or retrieve a clean record of who approved it.
The insight we build on
Much of the industry is racing to connect agents: messaging between them, access to tools, rails to move money. Those are problems of interoperability, and they are being solved. The gap Subchain addresses is different. It is fiduciary accountability across a company boundary — an enforceable point at which an agent-negotiated proposal becomes a commitment a named person has approved. Whoever owns the contracting lifecycle and its record, not merely the transport, sits where liability and trust are actually decided.
Why now
No enforceable, agent-specific contracting standard exists yet, and the frameworks enterprises rely on were written before autonomous, tool-calling agents. The legal foundation, though, is already in place: the UNCITRAL Convention on the Use of Electronic Communications in International Contracts recognised, in 2005, that a contract can be formed by automated systems. The primitive is settled; the governance around it is not. Subchain is built for that gap — for enterprises already running agents in external procurement, in jurisdictions where automated contract formation is enforceable but agent authority, approval-gating and liability allocation remain unresolved.
How we are built
An AI-born company, governed by named people.
Subchain is designed from the outset to be operated by AI colleagues under human governance — the same division of labour the product enforces.
AI colleagues run the runtime
They conduct offer-counteroffer exchanges within pre-set authority limits, maintain and version the term-template library, run inline compliance checks and assemble the record — then surface exceptions rather than resolving them.
Named humans hold authority
Human governors set authority limits, thresholds and approval routing. When a proposal exceeds a configured threshold, the runtime halts and escalates for explicit approval before anything can bind.
Infrastructure, not an application
We chose to build a shared layer that sits between companies rather than a workflow for any one of them. That neutrality is the point: a standard is only trusted when it does not favour a participant at the table.
A DIFC-domiciled venture
Subchain is a Cohort Ventures venture, domiciled in the Dubai International Financial Centre, so that its approval gates and its record are built for a jurisdiction where governance and dispute resolution are already established.
What people ask before they engage.
What we are honest about.
The venture is early and has no named customers yet. We are also direct about the central open question. The pain we have observed is integration pain — the recurring cost of wiring each counterparty by hand and maintaining brittle one-off connectors. Demand for a governed-negotiation layer specifically is, at this stage, inferred from that adjacent pain rather than evidenced by signed design partners or buyer conversations. Turning that inference into evidence, by working with the first design partners, is the work in front of us. We would rather state that plainly than imply a proof we do not yet hold.
Where to look next
If you want the mechanism in detail, the product page describes the negotiation lifecycle and the approval gate step by step. If you are reviewing controls, the security page states plainly what protects this site today and what it does not yet claim.
Start a conversation with the team.
If your agents are beginning to commit to outside counterparties, we would like to hear how you govern that today — and whether a design partnership fits.