June 17, 2024 · 1 min read
Why Agents Need Tool Contracts
Abstract. A one-minute note on why tool contracts matter more than giving an agent more tools.
Giving an agent more tools does not necessarily make it more capable.
It often makes the system harder to reason about.
The useful question is not how many tools the model can call. The useful question is whether each tool has a clear contract:
- what authority does it have?
- can it mutate external state?
- is it idempotent?
- what errors can it return?
- what should the agent do after failure?
- what evidence proves that the tool succeeded?
A tool without a contract is not a capability. It is an uncontrolled side effect.
Good tool contracts make agent behavior less magical and more inspectable. That is usually the tradeoff worth making.