Where the two diverge
Both classes read a request and produce a response. The divergence is the last inch: an assistant hands the work back to a person, while operational AI finishes it and reports what it did.
That last inch is where nearly all of the time savings live, because copying a draft into another system is the repetitive part.
What operational AI requires
- A real system of record to act on.
- Callable operations rather than screens.
- Identity: it must act as a specific user in a specific organization.
- Confirmation gates for anything irreversible or costly.
- A complete audit trail of every executed action.
How to evaluate a claim
Ask for one end-to-end task and then look at the record. If a document exists, an appointment moved or an invoice was created — and there is an audit entry naming the actor — it is operational. If the answer is a well-written paragraph, it is an assistant.