Open Rails for the Agentic Economy
Open Rails for the Agentic Economy
Consider a plausible near future. Your AI agent handles procurement for your business, negotiating with supplier agents, comparing prices, executing purchases, settling invoices, hundreds of times per day. Now consider that every one of those transactions settles through infrastructure controlled by a single company. That company sets the fee on each transaction. It decides which agents are licensed to operate on its rails. It can freeze any participant's access with a policy update, for any reason, at any time.
You have seen this before. Mobile app stores take up to 30% of every transaction, decide which applications can exist, and have removed software from their platforms under commercial or political pressure. Payment processors have frozen accounts and reversed settlements based on internal risk models that no merchant can audit. These dynamics emerged when the transacting parties were humans choosing platforms deliberately.
When the transacting parties are autonomous agents executing thousands of interactions per hour, platform-controlled rails become something worse than expensive. They become a structural vulnerability. An agent cannot pause mid-execution to renegotiate terms, switch providers, or appeal a frozen account. It simply stops working. And every agent downstream of it stops working too.
What does open infrastructure mean?
Here is what it means in concrete terms, and why it matters specifically for agent commerce.
Open settlement infrastructure requires that no single entity controls who can participate. Any agent, any principal, any merchant can transact by presenting a valid key pair. There is no application process, no licensing agreement, no approval queue. The rules governing valid transactions are enforced by protocol consensus across thousands of independent operators, and those rules can only change through an onchain governance process that is itself auditable and open to anyone with a stake.
MultiversX was built with these properties from its origin. Around 3,200 independent validators operate the network today, distributed across three execution shards and a coordination metachain. Since the Andromeda upgrade in May 2025, all ~400 validators in each shard sign every block. This is a specific and consequential design choice: it means no subset of validators can unilaterally alter the record. Producing a fraudulent block would require compromising a supermajority of an entire shard's validator set simultaneously, and up to a third of each shard's validators are reshuffled every 24 hours.
The sharded architecture matters here for a reason beyond throughput. In a monolithic chain, every validator processes every transaction, which means every validator holds the full state. That creates a single, complete dataset that a determined actor could target. In MultiversX's design, state is partitioned across shards. No single node holds the full picture. Capture of any individual shard's state does not compromise the network, and the metachain coordination layer ensures cross-shard consistency without consolidating state in one place.
Governance reinforces this. Since the Barnard upgrade in July 2025, protocol changes require onchain proposals with a 500 EGLD bond, a public discussion period, and a community vote. The rules of the system are changed by the participants of the system, through a process anyone can observe.
A question of speed
There is a counterargument that matters, and it is worth addressing directly: open infrastructure has historically been slow. If the agentic economy requires sub-second settlement and open rails cannot deliver it, then closed platforms win by default, regardless of their structural risks. Speed is a prerequisite, not a feature.
This is precisely what the upcoming Supernova upgrade delivers. Block times drop from six seconds to 600 milliseconds. Intra-shard finality has been measured at under 200ms in testing. Cross-shard settlement completes in under two seconds. The mechanism that enables this is the decoupling of consensus from execution: validators agree on transaction ordering first, and execution follows in parallel across shards. This is the architectural payoff of building a sharded system from the start. Consensus and execution can be separated cleanly because the shard structure already partitions work across independent processing groups.
A monolithic chain that wants sub-second finality has to make every validator do everything faster. A sharded chain distributes the work and lets each group operate at its own optimal speed. At agentic scale, that difference is the difference between a blockchain that holds and one that doesn't.
Future-proof architecture
The agentic economy will be built on settlement infrastructure. The question is what kind. The best version of that infrastructure is open, permissionless, and governed by its participants. It performs at agent speed. It scales with demand. And it ensures that the rails themselves remain a public good rather than a private tollbooth.
That is the infrastructure MultiversX is building: fast enough for real-time agent commerce, secure enough that no single party can alter the record, and open to anyone with a key pair.
This is what Supernova is for.
