“We want an AI agent” is now the opening line of half my discovery calls. About 70% of the time, what the client needs is an automation — maybe with one AI step inside it. The distinction isn’t pedantry; mixing them up produces systems that are either fragile or pointlessly expensive.
Automations follow paths
An automation is a fixed route: trigger, steps, outcome. Form submitted → contact created → notification sent. It does exactly what it did yesterday, which is precisely its value. When an automation fails, it fails loudly and mechanically — a webhook didn’t fire, a field was missing.
Most business workflows are this. Data moving between tools, notifications, syncs, scheduled reports. They need reliability engineering — retries, error routing, idempotency — not intelligence.
Agents choose paths
An agent is given a goal and tools, and it decides the route: which tool to call, what to do with the result, whether to try again. That autonomy is powerful exactly where inputs are too varied for rules — and dangerous everywhere else, because the failure modes are subtle. An agent doesn’t crash; it does something plausible and wrong.
That’s why every agent I ship carries three things an automation doesn’t need: an eval suite that measures behavior on real historical cases, confidence thresholds that route uncertain cases to a person, and logging detailed enough to answer “why did it do that?”
The hybrid is usually the answer
The pattern that actually works in client operations: a deterministic automation as the spine, with agent steps only at the junctions that need judgment. A sales-research system I built runs a fixed pipeline — trigger, gather, log — but the research and drafting in the middle is agentic, and a rep approves before anything sends.
If a vendor can’t tell you which parts of their “agent” are deterministic and which parts decide, that’s the tell. The teams that ship reliable AI systems know exactly where the dice are rolled.
Related service: AI Agents & Internal Assistants · Proof: AI sales research and outbound agent for a 50-rep B2B team