Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

Agentic AI hype vs. RPA reality: Adaptation isn’t so simple

Agentic AI has the potential lifecycle to upgrade automation dramatically, but I'm not convinced RPA vendors are paving the way to use it effectively.

Agentic AI hype vs. RPA reality: Adaptation isn’t so simple

[Song_about_summer / Adobe Stock]

Profile photo of Mukesh Kumar

I’ve been part of the automation world since the early 2000s, and there’s been a fascinating evolution from clunky old screen-scraping tools to the current wave of “intelligent process automation” claims. While agentic AI excites me with its potential to redefine automation, I’ve got my reservations about whether robotic process automation (RPA) vendors are ready for the transition. Here’s why.

RPA’S CURRENT CHALLENGES

RPA has come a long way, but some core issues are more prominent than ever:

1. Fragile Systems: RPA bots have a troubling history of falling apart over the smallest changes—a new UI or data field is often enough to send configuration into disarray. I’ve seen companies pour money into keeping these bots afloat through endless maintenance.

2. Understanding Issues: While RPA bots can process volumes of data, understanding context is another ball game. When faced with something like a vendor invoice missing crucial details, they can make errant assumptions or simply stall.

3. Handling Unstructured Data: For all the claims, most RPA platforms struggle with unstructured data types like emails or images. I’ve witnessed firsthand how projects stumble due to exceptions that genuinely require human intervention.

Agentic AI can potentially fix these problems with tools like LLMs that address ambiguity and adapt easily. The real question, though, is whether vendors can successfully pivot to embrace this change.

WHERE VENDORS GET STUCK

Vendors aiming to leap into the Agentic AI world are frequently tripped up by some shortcuts:

• Old Infrastructure with New Tech: A lot of these solutions bolt-on LLM calls onto their existing automation technology. Pairing legacy technology leads to complex solutions that miss the whole point of context-aware agents with a lot of autonomy.

• Mistaking Orchestration for Intelligence: The whole promise of AI hinges on quality data. Just relabeling platforms doesn’t mend fundamental issues like varying data quality or disconnected APIs that impede successful functionality.

• Generic Models: While generic models are great for some domain use case like marketing, more intelligence is needed for accurate responses in domain-specific applications. It is going to be hard for the RPA vendors to have breadth of industry models.

PRAGMATIC STEPS FOR PROGRESS

The vendors that thrive will be those taking realistic actions, rather than selling dreams of all-functioning AI:

• Hybrid Approaches: AI can help, but human judgment should be at the ready for complex exceptions. I’ve seen a logistics company cut errors by 60% this way.

• Specific AI Development: The broad claims must give way to partnerships with clients that build AI agents working best with tailored, well-understood data.

• Integrated Ethics: Transparent AI that explains its actions and mitigates bias should be integral, not an afterthought requiring additional purchases.

CONCLUSION

Agentic AI has the potential lifecycle to upgrade automation dramatically, but I’m not convinced RPA vendors are paving the way to use it effectively. Those who thrive will do so by acknowledging their strengths and limits, emphasizing human collaboration and pursuing genuine intelligence instead of trending buzzwords.