Last month leaders at UiPath invited me to attend the company’s Forward conference after reading my article in Forbes about AI agents. The conversations that arose led to some interesting insights about agentic AI applications, which are attracting more attention for enterprise uses by the day.
For those of you who don’t know about UiPath, the company makes a technology platform that historically focused on robotic process automation. RPA (which doesn’t usually involve mechanical robots) is a technique that helps digitize highly structured business processes. When it’s implemented properly, RPA can lead to increased accuracy and speed and decreased costs, among other benefits. Often RPA is used for tasks most humans would not necessarily want to do all day long. A great example of RPA is the digitization of paper-based documents (say, checks issued by banks) into digital records or transactions (e.g., updating financial and banking backends). In fact, UiPath has a significant base of intellectual property in the machine learning and document processing functions necessary for RPA.
A New Agentic Strategy Announced At UiPath Event
The main focus of Forward was what CEO Daniel Dines called Chapter 3.0 of the organization, which will more heavily focus on agentic applications as well as RPA. It’s an interesting direction, given that many demonstrations of AI agents from the likes of Salesforce and ServiceNow suggest that agents may take the place of RPA. But after having a conversation with UiPath CTO Raghu Malpani, I am convinced that UiPath is onto something here.
At the conference, UiPath announced a number of new capabilities to extend its RPA Platform. But the key was the belief that AI and agentic programming are an extension of the existing platform. This is not unlike what we are seeing from ServiceNow or Salesforce. But my takeaway was that UiPath is different, in that its agent approach has a higher focus on enterprisewide agents versus the more personal or departmental approach we see in some other platforms. This viewpoint comes from three new capabilities.
First, Agent Builder provides a more contextual and guided development experience. It is still low-code like other tools in this space, but there is a richer set of controls to guide the process. Think of it like a really good form to be filled out versus a blank sheet of paper. This context gives developers just enough structure for the supported workflows, integration points and access controls.
Second, Connector Builder helps with scale. The key to scaling agents will be similar to other non-AI applications in that the agent will need secure access to high-scale, cross-platform integration points. Connector Builder provides developers with an API-based integration hub that will enable a governed and manageable service for agentic applications.
Third, Autopilot for Everyone is a way for business users to create their own personal or departmental agents. One new concept of agentic programming that is not getting a lot of attention is that agents can use other agents or RPAs. This means more of a workflow assembly experience than a traditional process of application development. Too many platforms are treating the high-scale and personal agents separately, but I think that personal agents are just another onramp to get to enterprise agents. UiPath Autopilot provides users with the ability to create a personal agent that could invoke an enterprise agent as needed.
Complementary, Not Competitive
Over the past few months, we have seen many demos of an AI agent facilitating a business workflow, with the large language model behind it making most of the decisions along the way. This approach unquestionably has some utility, especially for processes with a degree of ambiguity such as granting an exception to pre-existing rules. It is also helpful for low-code and no-code development since some of the agents’ logic is delegated to the LLM—which makes agents easier to create and deploy.
However, LLMs are not deterministic by nature. They tend to do better with ambiguity rather than set business rules. By contrast, RPA processes are built with the expectation of determinism. So, both types of applications will remain necessary for businesses, especially for higher-scale processes. In fact, an AI agent could invoke an RPA process—or vice versa—in cases that need both determinism and less-structured decisions, such as mortgage approval.
An Inspired And Motivated User Base
The reaction to UiPath’s announcements at the event was excitement. I spoke with multiple users who were considering how adding agents into their existing deployments could add value quickly. However, one of the more interesting conversations was with a user with an existing RPA that generated many exceptions requiring human handling. Her sense was that most of these exceptions were relatively simple to resolve, and that by adding an agent to handle the easier RPA exceptions, she could further reduce human touches. So, as I mentioned earlier, this is a case where an agent and an RPA could work collaboratively to increase velocity and quality.
The Value Of A Shared Platform
If we accept that both styles of process applications will remain viable for some time, we must also consider whether there is a way they can share services or infrastructure. This is where UiPath makes a solid case. Security, governance, workflow management and monitoring are all capabilities that have been in place within its RPA platform for some time. A unified RPA and agentic platform may appeal to enterprises that worry about issues such as application sprawl, increased attack surfaces and maximizing reuse. Importantly, these are all areas in which other platforms or general-purpose development tooling offerings that support agentic applications have caused some concern.
This is great news if you are an existing UiPath customer, since it will likely be easier for you to deploy and manage AI agents. Clearly, the existing users at the conference understood this. However, if you are not an existing customer, the new functionality does raise the same questions as other platforms that embed agentic capabilities such as Salesforce. For instance, will it require too much work or cost to deploy an entirely new platform only for creating agents—as opposed to simply building agents using general-purpose toolsets from one of the AI-aware hyperscalers such as AWS or IBM?
Keeping What Works, Augmenting With What’s New
As someone who spends a great deal of time explaining new frontiers in AI application development, I was quite taken by the pragmatism of both UiPath and its users. All too often, new technology is represented as a replacement for something that already does a pretty good job as it is. UiPath’s approach, however, is to use something new (agents) to augment something that’s been working well for a long time (RPA). This means that agents are not regarded as either a threat or a panacea, but rather a catalyst to better apps and better enterprise processes.
For existing UiPath customers, getting started will be straightforward. However, UiPath may need to consider investing in more market education or deploying more aggressive sales methods to help new customers get on board. And that may be a great investment since UiPath’s vision speaks to both business value—starting in the short term—and to process evolution versus a disruptive revolution.
