RPA robots follow a fixed script exactly the same way every time, which makes them predictable and easy to audit. AI agents reason about a goal, handle unstructured input and exceptions, but cost more and won't always decide the same way twice. In regulated businesses the best designs use agents for judgement and robots for execution.
George McKenna, Chairman and CTO, ETT Group. Updated 15 October 2026.
| RPA robot | AI agent | |
|---|---|---|
| Good at | Repeating a defined task exactly | Reading messy input, deciding, handling exceptions |
| Predictability | Same result every time | Can vary between runs |
| Cost per transaction | Low | Higher |
| Works with legacy screens | Yes | Usually needs an API, or a robot to act for it |
| Easy to audit | Very | Only if evidence is designed in |
The pattern that works: the agent handles judgement, the robot carries out the decision in the system of record through a tested, permissioned path, a person approves where the risk calls for it, and an orchestration layer records every step. The agent never touches core systems directly.
If you already run RPA, you have the hardest part built. An agent in front of your robots can take on the reading and deciding that still lands on people today.
Related questions
Show Your Working
A fortnightly piece on building and buying AI that pays its way. What I'm building, what it costs, and the questions to ask before you sign anything.