I’ve been following Newton Protocol’s marketplace design for a while and one thing keeps coming to me. As soon as agents can compose services with each other, some of them will try to pay each other for an edge.
From what I understand, Newton is built around four participants. Developers publish agents to the model registry. Operators stake NEWT and compete to run those agents and execute tasks. Users submit intents. Validators secure the network. Every task has to come with ZK proofs and operators get slashed if they don’t deliver. Operators also build reputation over time based on how reliably they execute.
Newton’s roadmap talks about moving into agent-to-agent workflows and service composition. That’s where I see the shift. Right now it’s mostly user to agent. But soon it’ll be agent A to agent B to agent C. Like a strategy agent calling a price oracle, then a swap agent, then a yield agent.
Once that happens, I think incentives will get messy. Newton already uses an EIP-1559 style fee market for ordering and operators already compete on speed to win work. If agents themselves can hold NEWT, it wouldn’t surprise me if they start offering small payments to each other to get picked first.
In my view it could look pretty simple. A price oracle agent gets two requests. An arbitrage agent offers it 0.01 NEWT to be called first. A routing agent offers a tiny kickback to be included in the workflow. Is that wrong? Not technically. It’s just market behavior. But it does mean the workflow stops being only about the user’s intent. Now it’s also about side deals between agents.
The zkPermissions system today does a good job of controlling what an agent can do with my assets. But I don’t see anything yet that governs what Agent A pays Agent B. And the protocol does punish dishonest execution, but paying for priority isn’t called out as dishonest.
So my take is that Newton will probably need to think about an Agent Interaction Policy. I’m not saying ban all payments between agents. I’m saying make it transparent.
If I were designing it, I’d do three things. First, require any agent-to-agent payment during a workflow to be declared upfront in the intent. If it’s hidden, the proof should fail. Second, let users set allow_agent_payments to false in zkPermissions for workflows where fairness matters more than speed. Third, add this to operator reputation. If you facilitate off-book payments, your score drops.
I don’t think this kills composition. I think it protects it. Because if users can’t trust that the workflow follows their intent, they won’t use it.
Newton wants to be the coordination layer for onchain automation. For that to work long term, I believe we need rules not just for execution, but for how agents interact with each other.
This is just my analysis based on how the marketplace and fee market are structured today. But I think this is a conversation we should have now, before agent bribery becomes the default. NFA.DYOR.