Agentic AI and the Future of Robotic Process Automation

By Jeff Richardson, Director, Data Strategy and Automation, Vantage West Credit Union

Recent advancements in agentic AI have created a significant overlap in use cases traditionally reserved for Robotic Process Automation (RPA). Both technologies excel in multi-step workflows that require API integration and external data access to reach a specific objective. Specifically, they are both ideal for processes that are manual, repetitive, and cross-functional. These target processes typically share three key traits:

• High Volume: Tasks that are frequent and repetitive.
• Complexity: Workflows involving multiple distinct steps.
• Interoperability: Procedures that span multiple legacy systems.

Now, clearly, there are some significant differences, most notably agentic AI’s ability to move beyond the rules-based approach of traditional RPA to use reasoning and contextual understanding to make autonomous decisions even when complex goals exist. In light of this, it is natural to wonder if agentic AI will supplant RPA and relegate it to an ancillary role within a company’s tech toolbox.

While we can’t be sure of where RPA will land in the long-term, right now we are seeing some use of RPA in conjunction with agentic AI that suggests that RPA will continue to be used, albeit in a somewhat different fashion than has been the case in the past. I know of multiple instances where agentic AI is being used as an orchestration layer that utilizes its reasoning to break a complex request into discrete steps and then passes these to RPA to execute.  While RPA is strictly rules-based, agentic AI introduces reasoning and contextual understanding to navigate complex goals. Before the integration of agentic AI, automating a process required exhaustive documentation and scripting of every possible logic path. If the system encountered an unscripted exception, it resulted in immediate failure or required manual intervention. In contrast, agentic AI utilizes dynamic learning to self-correct, creating a far more resilient and robust automation solution.

Leveraging agentic AI within your automation space can bring significant benefits to those who do so prudently.

The marriage of RPA with agentic AI also enables automation to tackle some of the knottier problems that have been the bane of RPA’s existence.  A prominent example is unstructured data such as handwriting.  For instance, while standard Optical Character Recognition (OCR) is effective for printed text, it often fails to accurately digitize handwriting.  Agentic Document Extraction (ADE) addresses this gap, demonstrating remarkable accuracy in interpreting handwritten documents through contextual reasoning.

Some of these same RPA tools are now incorporating Natural Language Generation (NLG), which is AI that allows users to describe what they want the agent to do as opposed to coding/scripting the particular workflow needed.  Moreover, AI can be used for human-in-the-loop automations by allowing users to initiate automations, validate results, aid in handling exceptions, or make key decisions via chat or other means involving natural language processing (NLP).  A major advantage of AI as opposed to traditional attended automations of this type is that AI can learn from human interaction and improve over time.

It’s already the case that RPA software companies such as UiPath are embracing precisely this model with the rollout of Maestro, which is its agentic orchestration engine.  Automation Anywhere now allows users to build their own agents to work in conjunction with their RPA bots.  Likewise, Microsoft Power Automate and Copilot Studio enable agent building and automation within the Microsoft suite of products.  UiPath has also recently acquired WorkFusion to leverage the pre-built agents that WorkFusion possessed and further advance UiPath’s shift towards agentic AI. 

To bring this together using an example from the financial industry, consider a loan application review: An AI agent acts as the orchestrator, gathering various structured and unstructured documents, such as tax statements and identification. If the agent identifies variances, it attempts to resolve them within the institution’s established guidelines. Once the data is verified, the agent hands off the materials to an RPA bot for high-speed entry into the core banking systems. This creates a tripartite partnership—AI for reasoning, RPA for execution, and humans for final underwriting—to drastically reduce processing times and error rates.

I think that what we will increasingly see is not the wholesale elimination of RPA but instead the hybrid model that has been described above.  Nevertheless, legacy, stand-alone RPA bots that have performed well on repetitive tasks over the years do not necessarily need to be replaced and many will choose to leave them running unchanged. 

Leveraging agentic AI within your automation space can bring significant benefits to those who do so prudently. Since the cost model (consumption-based for many AI vs. licensed-based for RPA) may differ and automation capabilities are now greatly expanded, companies must have a framework for evaluating the approach of future automations as well as determining the utility of changing the flow of existing automations.  For organizations new to agentic AI, a prudent starting point is to launch one or two carefully chosen pilot programs that are most likely to yield meaningful, measurable results. Everyone who wants to get the most out of their automation program needs to pay attention and follow the path of agentic AI because of the potential for increased efficiency and cost reduction, further reduction in errors, and greatly expanded capabilities while leveraging existing RPA investments.