Fallback
A backup path used when the first choice fails, is unavailable or is not confident — such as a different model, a simpler rule or a person. A good fallback is designed and tested, not an afterthought.
A fallback is what happens when the primary path does not work. In AI systems it can mean routing to another model when the first is slow or down, switching to a simpler rule, retrying with different settings, or handing the case to a person.
Fallbacks improve resilience but complicate evaluation. If a fraction of requests silently fall back to a different model, the quality and cost you measured may not describe what customers get. That is why routing metadata, which shows which model actually answered and whether a fallback occurred, is becoming evidence, and why fallback conditions belong in the runtime address and the serving route.
Decision systems
In a decision system, a fallback is not the same as abstention. Deferring an uncertain case to a person is a deliberate design choice, not a failure mode: uncertainty needs somewhere to go. Test fallbacks as part of the unhappy path, and note how they change model routing.
Read more in The benchmark needs a runtime address.
Related terms
Model routing
Directing each AI request to a particular model or route based on cost, quality, latency or availability, so the model that answers can differ from request to request.
Serving route
The actual path a request takes to be answered — which model, hardware, precision, routing and fallback rules handled it — as opposed to the model name a vendor advertises.
Abstention
A deliberate option for an AI decision system to decline to decide and return uncertain or high-consequence cases to a person, rather than forcing every input into yes or no.
Used in these essays
OpenAI Decisions API turns probability into policy
The provocative part of OpenAI's Decisions API is not that AI can make choices. It is that a model score can quietly become an action — and an action can quietly become policy.
The hidden metric in AI automation is supervision
As AI moves from assisting work to leading it, hours saved stop telling the whole story. The scarce resource shifts to human supervision: approvals, exceptions, context and judgment.
The benchmark needs a runtime address
AI systems are increasingly dynamic at runtime. Enterprise evaluation should qualify the serving route, harness, tools and fallback conditions, not just the model name.