All articles
1 min read
AI features worth building (and the ones that are not)
A practical filter for deciding where a model belongs in your product — and where a query would do.
AIProductEngineering
The filter
Before adding a model, ask: is the task ambiguous, high volume, and tolerant of being roughly right? If any answer is no, write a query instead.
Where it works
- Summarising long unstructured input
- Classifying inbound requests
- Drafting a first version a human edits
Where it does not
Anything where a wrong answer is expensive and silent. Financial calculations, compliance checks, permission decisions.
Cost as a design constraint
Token cost is a product decision, not an infrastructure detail. Cache aggressively, and pick the smallest model that passes your evaluation set.
