Blast radius first
What the agent may touch, what it may never touch, and the maximum single-transaction and per-hour loss if every guard fails. Written down and agreed before a line of agent code exists.
An agent that drafts an email can be wrong. An agent that signs a transaction, rebalances a position or releases a payout cannot. The mistake is irreversible, on a public ledger, and denominated in your customer's funds. That is a different engineering discipline from prompt design.
We build agents for systems where being wrong costs money: simulated before execution, scoped to permission boundaries, bounded by spend ceilings, and logged so every decision can be reconstructed months later.
Two records, two people. One of us writes protocol integration and execution-simulation code in Rust. The other ran payment infrastructure at $60M+ volume in MAS-regulated, SOC 2 / ISO 27001 environments.
We do not sell an agent on its own. Agent work is scoped inside the execution, risk or settlement path it touches, because that is the only place its blast radius can be bounded.
Scope
What the agent may touch, what it may never touch, and the maximum single-transaction and per-hour loss if every guard fails. Written down and agreed before a line of agent code exists.
Actions are simulated against forked state and evaluated before they are signed. Deterministic steps stay deterministic. Models are used for judgment, never for arithmetic or as the system of record.
Signing isolated behind a permission boundary, per-action and per-day value ceilings, and a stop your team can reach. Key custody designed with you, never improvised.
Every decision logged with the inputs and the simulation that justified it, so an auditor, a counterparty or your own risk team can reconstruct why the agent did what it did.
Founder · replies same business day
Fixed fee, scoped in writing before anything starts. No hourly billing, no verbal scope.
Related: Trading & execution infrastructure · DeFi protocol engineering · How we work & pricing