Robotic process automation repeats exact steps reliably. Language models cope with variation in text and judgement. Using the wrong one produces either brittle scripts or expensive, unpredictable steps.
Where each fits
- Rules or RPA — fixed screens, structured data, deterministic decisions, high volume, audit needs.
- AI — unstructured inputs such as emails and documents, classification, extraction, summarization and drafting.
- Both — AI reads and structures the input; rules validate and execute the action.
A good pattern
Let the model produce structured output, validate it with deterministic checks, route low-confidence cases to a person, and perform the final action with ordinary code. The model handles ambiguity; the rules handle consequences.
If a step can be written as a clear rule, write the rule. Use a model only where the variation is real.