Every martech vendor relabelled their rules engine as agentic in the last eighteen months, and teams now reach for agents on tasks a three-line rule would handle. There is a fast test.
The test
Can you write the decision as one sentence beginning with if? If yes, write the rule. An agent adds cost, unpredictability and an evaluation burden for nothing.
If the honest answer is it depends on about nine things and I would have to look at each case, that is where an agent earns its place.
Worked examples
- Send a welcome email on signup. One sentence. Rule.
- Route a lead to the right salesperson by territory. One sentence. Rule.
- Score an inbound lead against everything we know about past closed-won deals. Not one sentence. Agent.
- Summarise a sales call into structured CRM fields. Not one sentence. Agent.
- Suppress a contact who unsubscribed. One sentence, and a compliance obligation. Rule, and make it synchronous.
Why this matters beyond cost
Rules fail predictably and visibly. You can point at the branch that broke. Agents fail unpredictably and quietly, because a bad decision looks exactly like a good one until someone inspects the outcome.
That difference is why compliance-adjacent logic should never sit behind an agent. Consent suppression, retention deletion and unsubscribe handling want deterministic rules you can audit, not a model deciding case by case.
The middle ground worth knowing
Many tasks split cleanly: an agent does the judgement and a rule does the action. The agent classifies intent, the rule decides what happens for each class. You get the flexibility where variation is real and determinism where consequences are.
General information, not legal advice. Rules current as of July 2026.