Agentic automation

Agentic automation is automation in which software decides the steps needed to reach a stated goal, rather than following a fixed sequence defined in advance by the person who configured it.

The contrast is with rule-based automation, where a trigger produces a predetermined series of actions. Rule-based automation fails visibly when reality does not match the rule. Agentic automation adapts, which is its advantage and the entire source of its risk.

Because the steps are chosen at run time, they cannot be fully reviewed in advance. That is why agentic systems doing anything consequential need an approval gate: the boundary has to be placed around outcomes, not around steps that nobody has seen yet.

The practical test for whether something is agentic is simple. If you can print the complete list of actions before it runs, it is rule-based. If you can only print it afterwards, it is agentic, and it needs a different kind of supervision.

See also human-in-the-loop and agent workforce.