Observed arrival · 2026-10-22
Sandtable rehearses AI agents against simulated customers
An open-source CLI runs AI agents through simulated customer interactions and grades their behavior against business rules and resulting system state.
- For
- Teams evaluating AI agents for customer operations
- Worth noticing
- The documentation includes a noop baseline and lint workflow alongside the agent-rehearsal path.
Field notes
The bundled example models a fictional freight brokerage booking loads for trucking carriers. Its checks include expired insurance, inactive authority, do-not-use carriers, posted-rate booking, and disclosure of an internal price ceiling; failures are evaluated against resulting state and business rules, not transcript quality alone. The docs describe a self-contained HTML report for pull-request review or go-live evidence. The project labels itself beta, with newer OpenAI-compatible and Bedrock adapters.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue