Judgment up front, code behind, a person when it is unclear
AIMade by AISummarised by AI (Claude) from one piece of material it read (a talk or an explainer).How it's made
When a decision is automated, what to leave to the model and what code should hold comes up again and again across several write-ups. The starting point is that even a single inquiry saying a card was charged twice and asking for a refund needs small judgments first: who it is assigned to, what the request actually means, and whether it can be handled automatically. The division called most practical has four branches: a small model reads the meaning, code compares the payment record against the refund rules and splits the flow, a large model writes the answer only at the moment long prose is needed, and unclear cases go to a person. A threshold is not simply picked; it can only be set after checking on your own data that predictions that come out at 90% really are right about 90% of the time. The core of this structure is the line drawn between saying there is a 99% probability that this is a refund request and saying the customer is entitled to a refund.
Key points
- Even one inquiry needs small judgments first: assignment, intent, and whether it can be handled automatically
- The model reads the meaning while code compares the rules and splits the flow
- A threshold goes in only after checking on your own data that the probabilities hold
- A large model is called only at the moment long prose is needed
This is a note: a summary an AI (Claude) wrote of one piece of material it read (a talk or an explainer). It has no source link; events checked against a source are kept as records.