Workflow automation
An AI-assisted system whose path from input to output is knowable in advance - model calls are embedded at specific decision points inside otherwise deterministic code.
A workflow automation is the appropriate architecture when the path from input to output is predictable: invoice received → classify → route → approve under a threshold → notify. The logic is explicit; the model, if used, is used for one specific step (extraction, classification, short-form generation), and the rest is deterministic code.
Why it's usually the right default
Automations are cheap to run (no reasoning-tier model, minimal tokens), easy to monitor (each step is explicit), and straightforward to governance-audit (the logic is readable code). They break in boring, predictable ways - a schema change, a rate limit, a queued retry that wasn't idempotent - which is exactly the kind of failure mode a mature engineering team handles well.
When to pick an agent instead
When the path from input to output cannot be written down in advance. When the system has to decide, per invocation, which tools to call and in what order. See AI agent and the longer treatment in AI agents vs automation.
The hybrid pattern
The best production systems are neither pure agents nor pure automations. They are automations with a small number of delegated sub-tasks where an agent is genuinely appropriate - a claims workflow, for example, that uses an agent only for "research and summarise the disputed case for the human reviewer" and keeps the rest deterministic. This is what well-designed production systems usually converge to.