Guides
How to build an agent that keeps its money.
Practical guides for teams whose AI agents can move money, from refunds and discounts to payment plans and negotiated terms: how to test them, which guardrails stop the common attacks, and how the major agent platforms handle money actions.
Last updated
Testing and guardrails
- How to test an AI agent that can issue refundsStep-by-step method for testing AI refund agents: mock tools, policy as checkable clauses, adversarial scenarios, repeated runs, dollar scoring, release gates.
- Guardrails checklist for AI refund and discount agentsControls an AI refund or discount agent needs, what each stops and how to test it: server-side caps, approval IDs, claim counters, injection handling and more.
- How to write a refund policy an AI agent can follow and a test can checkTurn refund and discount SOPs into numbered clauses an AI agent can follow and a script can check: limits, approval tiers, never-items and required disclosures.
- AI red teaming vs adversarial testing vs simulation testingWhat jailbreak red teaming, cooperative simulation testing and money-focused adversarial testing each find in AI agents, what each misses, and when to run each.
- Why one test run is not enough: measuring agent reliability with pass^kHow to measure AI agent reliability: pass^k, failure rates with Wilson confidence intervals, paired comparisons for regressions, and how many runs you need.
- Indirect prompt injection against support and shopping agentsHow hidden instructions in tickets, attachments, product pages and tool output reach support and shopping agents, and defenses for refund and purchase tools.
By platform
- Intercom FinIntercom Fin and refunds: Procedures, guardrails and testingHow Intercom Fin issues refunds through Procedures and data connectors, which guardrails you can configure, and what Fin simulations and batch tests cover.
- ZendeskZendesk AI agents and refunds: actions, escalation and testingHow Zendesk AI agents process refunds through actions and API integrations, which escalation and integration controls exist, and what the test widget covers.
- Salesforce AgentforceSalesforce Agentforce and financial actions: guardrails and testingHow Salesforce Agentforce runs refunds and credits through actions and subagents, which guardrails the platform enforces, and what Testing Center checks.
- DecagonDecagon AI agent refunds: AOPs, guardrails and simulationsHow Decagon agents handle refunds through Agent Operating Procedures and tools, which guardrails Decagon describes, and what its simulations do and don’t cover.
- SierraSierra AI agent refunds: journeys, supervisors and simulationsHow Sierra agents handle refunds and returns through journeys and tools, what supervisors and the Agent SDK control, and what Sierra’s simulations test.
- OpenAI Agents SDKBuilding a refund agent with the OpenAI Agents SDK: tools, guardrails and approvalsHow to build a refund tool with the OpenAI Agents SDK: tool guardrails, approval for tool calls, conditional tools, tracing, and what you need to test yourself.
Free checklist
Get the guardrail checklist for agents that move money.
A one-page PDF of the controls and the test for each. Download it right away, and we email you the link.