The Standards Gap
We have standards for everything in traditional finance. Order types, settlement procedures, risk disclosures, audit requirements. These standards took decades to develop, usually after something went wrong.
Autonomous onchain finance doesn't have decades. Agents are already managing capital, executing trades, and interacting with protocols. But there's no shared understanding of what "safe execution" looks like.
What Needs Standardizing
Three areas need immediate attention:
1. Transaction Validation
Before an agent submits a transaction, what checks should it perform? At minimum:
- Simulation: Does the transaction succeed in a simulated environment?
- Slippage bounds: Is the expected execution price within acceptable range?
- Gas economics: Does the expected return justify the execution cost?
- State freshness: Are the inputs to this decision still current?
These seem obvious, but there's no standard way to express or enforce them.
2. Execution Receipts
After a transaction executes, what should the record contain? The raw transaction receipt isn't enough. We need:
- The decision context (why this trade, what strategy, what parameters)
- The pre-execution state snapshot
- The expected vs. actual outcome
- Any constraint violations or near-misses
This creates accountability even when there's no human to hold accountable.
3. Emergency Procedures
When something goes wrong, how should the system respond? Every autonomous system needs:
- Circuit breakers: Automatic pause when anomalous conditions are detected
- Graceful degradation: Fallback to a safe state, not a dead stop
- Recovery procedures: How to restart after an emergency halt
- Post-mortem data: Everything needed to understand what happened
A Practical Framework
I've been thinking about this as three concentric layers:
Must Have (safety-critical): Transaction simulation, position limits, emergency halt capability. Without these, the system shouldn't run.
Should Have (operational quality): Execution receipts, performance metrics, alerting. Without these, you're flying blind.
Nice to Have (ecosystem maturity): Cross-system interoperability, standardized reporting, formal verification. These come with ecosystem maturity.
The Path Forward
Standards don't emerge from committees - they emerge from practice. The teams building autonomous systems today are de facto writing the standards through their implementations.
The opportunity is to make this explicit. Share patterns, publish specifications, and build tooling that makes the right thing the easy thing.
The alternative - every team inventing their own safety model - doesn't scale. And the consequences of getting this wrong involve real money.