When we have this banner up at conferences, I get one of two responses:

“What the [pick an expletive] is an RPA?”

OR

“I hate my RPA.”

RPAs are probably one of the most incredible disappointing pieces of software.

When I say that, I don’t mean “incredibly disappointing”; I mean the most genuinely incredible in its promise and yet disappointing in its execution.

Robotic Process Automation is incredible because it tackles real problems that represent a significant operational cost. Because of this, at the end of even the most middling implementation of RPA solutions you will usually find a somewhat satisfied customer.

The disappointing part comes in the long tail of supporting and maintaining an RPA process.

Almost every RPA implementation you come across, no matter how good, is held up by a platoon of consultants, engineers, and subject matter experts tasked with constantly triaging the many ways the process might fail.

To what extent new, LLM-agent-driven “Cognitive Automation” ends up being a meaningful improvement over RPA remains to be seen. I think it’ll ultimately come down to whether the maintenance long tail is meaningfully better.