Use cases · headline
Act at 0.95. Call the reasoning model at 0.60. Ask a human below that.
Fast-and-slow AI systems need something that decides between the fast path and the slow one. Simo is the fast system and the decider, because its probabilities are calibrated.
The pattern
A fast system handles the routine, a slow one handles the hard, and something decides between them. Most stacks have no principled “something”: a prompt, a heuristic, or nothing.
- Clear case: Simo’s probability clears the act threshold, and the action executes on Simo’s answer alone.
- Ambiguous case: the probability lands in the middle, and the case escalates to a reasoning model. This is the only time the expensive call is made.
- Genuinely uncertain case: the probability falls below the ask threshold, and the case goes to a human with the probabilities shown.
Why Simo can be the gate
Because 0.8 means right about 80% of the time. No prose model gives you a number to put a threshold on. With calibrated probabilities, the thresholds are engineering parameters you can tune to the cost of an error.
What you save
Every case that clears the act threshold is a reasoning call that was not made. Count them: the savings are visible as a counter next to the gate.
Related
Read how calibrated probabilities work, and compare Simo with reasoning models.
Frequently asked questions
What thresholds should I start with?
Act at 0.95, call the reasoning model at 0.60, ask a human below that, then tune to the cost of an error in your product.
Does Simo replace the reasoning model?
No. Simo sits in front of it and handles the routine, so the reasoning model only sees the cases that need it.