Agent Failure Mode

A practical example

Before launch, a team lists failure modes for an outbound agent: duplicate emails to one contact, wrong tone for enterprise accounts, or writing to the wrong CRM field, each with a severity rating.

What to evaluate before investing

• Ask whether the vendor provides a documented failure mode catalog for the agent, or only generic error logs.
• Check what happens after each failure: does the run stop, retry, escalate, or continue with bad output?
• Confirm you can inject test cases that trigger known failure modes before the agent touches real leads.

Limitations and tradeoffs

Failure mapping is a snapshot; new failure modes appear when prompts, models, or connected data change, so the catalog needs ongoing maintenance.