What is a fallback agent?

A Fallback Agent is an AI agent that takes over when another model or workflow cannot confidently or successfully complete a task.

Fallback agents provide another path for handling situations that the primary AI system cannot resolve. Depending on the system, a fallback agent might gather additional context, use a different reasoning approach, route the task to another model, or escalate it to a human.

In Digits' Model Layer Cake, Fallback Agents help investigate novel or unclear accounting situations that the Company Model, Firm Model, and Global Model cannot confidently resolve. They can gather additional context and reason through the accounting work within defined guardrails. If the system still cannot reach a sufficiently confident result, the item can be escalated to an accountant for review.

Example: A transaction comes from an unfamiliar vendor with little useful information in its description. After the other model layers cannot confidently classify it, a Fallback Agent investigates the transaction and gathers additional context to help determine the appropriate accounting treatment.

Related terms: AI Agent, Company Model, Firm Model, Global Model, Model Layer Cake, Confidence Score, Review by Exception