Should coding agents fall back to another model when one fails?

I'd love some way for the prompt to work in a heirachy depending on complexity and confidence in the answer. Rather than wasting unnecessary expensive tokens on a task done cheaper.

I.e. Local Model - > Cloud cheaper cost model - > Frontier model.

So my prompt gives me the best value for money on any given input and output. This could then obviously have fallbacks similar to what the OP talks about.