When to Use an AI Agent vs. a Traditional Cron Job or Webhook

The most common mistake I see teams make with AI agents is reaching for one to solve a problem that a scheduled function or a webhook already solves — reliably, cheaply, and without the nondeterminism a model introduces.

A good rule of thumb: if the logic can be written as a fixed set of if/else rules, it doesn't need an agent. Agents earn their keep when a step genuinely requires judgment — interpreting unstructured input, deciding between several valid paths, or handling cases too varied to enumerate in advance.

This post is the framework I actually use with clients to sort their workflow backlog into "automate with a cron job," "automate with an agent," and "don't automate this yet."