LEVEL 03 · LESSON 26
Fallbacks
Plan the failure path as carefully as the happy path.
What it is
A fallback is a planned alternative when the preferred AI path cannot complete a task reliably or safely.
Why it matters
A useful system must handle uncertainty, unavailable services, invalid outputs, and missing information without making things up.
A business example
An order assistant cannot reach the shipping system. It explains that live tracking is unavailable and offers the normal tracking link instead of inventing a delivery date.
How it actually works
Identify failure conditions before launch and assign an appropriate response: clarify the request, retry a transient failure, use another service, return a limited answer, or route to a person. Bound retries by time and count. Preserve the information a human needs to continue. For actions with side effects, determine whether the first attempt succeeded before retrying, so an outage does not create duplicate orders or refunds.
A common misconception
A fallback is not always a bigger model. Sometimes the correct fallback is a clear ‘I cannot verify that’ with a useful next step.
ONE SENTENCE TO REMEMBER
Plan the failure path as carefully as the happy path.
CHECK YOUR UNDERSTANDING
An assistant cannot find a source that confirms a customer’s warranty status. What is a good fallback?
Show the answer
Answer: B.
Plan the failure path as carefully as the happy path. A fallback is not always a bigger model; sometimes the right response is a clear “I cannot verify that” with a useful next step. Bound retries and preserve what a human needs to continue.
Take it into your next meeting
What should your user see when the evidence source is unavailable?
One more concept, understood.
Mark this lesson complete to keep your place.