UiPath On Tour: Finding a balance between the Robotic and the Agentic?

last updated:

UiPath On Tour: Finding a balance between the Robotic and the Agentic?

last updated:

UiPath might have returned to a previous form of leadership, but it wants you to focus instead on how it wants its trusted robots to become active agents in your organization.

Last week, it was UiPath’s turn to bring their traveling roadshow to London, with “UiPath on Tour London” setting up in east London, just across the river from where Salesforce had parked their caravans last month. As I noted on that occasion, 2024 was becoming marked as the year where suddenly AI agents were everywhere, and this is a trend that is up and on its feet exceptionally fast; UiPath’s keynote sessions were not just heavily AI-focused but narrowed in, especially on that notion of agentic automation.   

On tour and shifting the line-up

UiPath’s 2024 so far has been somewhat challenging. It began with a new era of leadership, with Rob Enslin taking up the reins as sole CEO, having shared those responsibilities with co-founder Daniel Dines since he joined the company in mid-2022. In March, UiPath announced full-year (financial year 2024) numbers that saw it hit $1.308bn in revenue, yet by the end of May, Enslin had announced his departure from the company on the back of small miss against financial analyst’s expectations for the following quarter and a similar slight restatement of expectations for the full financial year. 

This meant Dines was back as sole CEO after a small spell focusing on products as Chief Innovation Officer (as well as serving as Executive Chairman of the Board). During his portion of the keynote session at “UiPath on Tour,” Dines obliquely noted, “Coming back as the CEO will help me bring back to UiPath, the fanatical customer centricity [..] where we’ve done everything for the sake of the customer”. It’s difficult not to wonder whether the organizational pains of growing so rapidly from a start-up into a billion-dollar-plus organization were mainly behind this set of events.

Robotic vs Agentic

Dines keynote was—as befits a CEO, albeit a reluctant one—setting the big picture for the direction of travel for UiPath over the next 3-5 years. Here, he plotted the evolution from the reliable, routine nature of the current generation of Robot Process Automation (RPA) operations, for which UiPath is a poster child, towards one where a broader range of more complex decision-making processes can be automated. 

Broadly speaking, RPA has been successful where tasks require identical, high volume, low exception processing. Where exceptions occur, they generally require human intervention, likely from a range of approved actions. Dines outlined a parallel set of intelligent operations – Agentic Process Automation (APA) – which is designed not to replace RPA but rather to be able to both design and operate more complex, decision-heavy tasks and processes backed by a generation of new AI models; Large Action Models (LAM). 

For LAMs, those previously referenced sets of approved human actions to resolve exceptions are examples of the sort of exemplars which form the basis of how they respond to a prompt asking for a next best action, in this case in reaction to a task exception. Whilst the concept of a “next best action” has formed part of the high-level AI use cases for decades, both APA and LAMs are at an extremely early phase of life and most of the research and development for the former has been focused on use with Large Language Models (LLMs) alone rather than in combination with LAMs (in short, because there is not yet a surfeit of those models to use, notwithstanding the hype around Rabbit earlier this year, and it’s somewhat unclear what will provide the vast library of actions required for their large scale development). 

What Dines is pointing toward here isn’t an unrealistic goal on the face of things. A form of automation that is in some way cognisant of the goal of the task, how it fits within a broader process, and able to suggest resolutions when it hits exceptions using a range of actions it believes will resolve them in a human-like way, using the intent of those actions within the overall goal as a guide. Comparing it as he did to self-driving in the sense that it’s more suitable for some roads than others suggests that this futurism is wedded to a degree of pragmatism. 

The Here and Now

UiPath CPO Graham Sheldon focused instead on the company’s deliverable short-term product cycle, where AI is part of a set of focuses along with the continued build-out of the platform toward a delivery path where cloud-based customers get first feature access. This includes ongoing work on the Process Intelligence and IDP elements that we’ve covered extensively. Practically, this means continued work to deliver specialized AI models for the day-to-day challenges that UiPath’s customers face: documents/data and tasks/processes. 

In the first instance, UiPath’s own Retrieval Augmented Generation (RAG) feature “Context Grounding,” was previewed, which has been developed with a focus on the sort of large, complex documents that tend to prove the most challenging and as such require specific tooling into the ingestion pipeline (e.g. document chunking) to be developed. The DocPath LLM – first announced along with that RAG feature at UiPath’s AI Summit this spring – that underpins the progress the company is making in IDP, was also highlighted. Autopilot for Developers (where an 80% acceptance rate for code suggestions was cited) and Testers is now generally available to customers, with a further set of features focused on the needs of Business Analysts set to follow.

Megatrends meet Megareality

The growth in the generation of automation platforms that bore UiPath was set against a pervading megatrend – Hyperautomation – that dominated discussions of how this technology fitted into enterprise operations for much of the previous decade. We were not fans of its tendency to over-promise a shiny chrome and glass future and its replacement as a staple of presentation decks with something agentic in flavor is not something that we especially relish, especially as it too doesn’t yet represent anything close to an implementable reality for almost any organization. As Alan noted on his visit to Automation Anywhere’s conference last month, on the ground RPA is still only scratching the surface in many of the organizations in which it has been adopted, with a huge amount of greenfield customers yet to be reached. Setting an ambitious direction of travel and providing reassurance that the company “gets it” technologically is probably necessary for UiPath in plotting its next phase, but much as Dines noted in the context of why he finds himself in a sole leadership role at the company again, it must not lose the on-the-ground reality of the customer-centricity that brought it this far.

Subscribe to our monthly newsletter

Leave a Comment

Subscribe to our monthly newsletter

We use cookies to ensure that we give you the best experience on our website. If you continue to use this site we will assume that you are happy with it.

Work Intelligence Market Analysis 2024-2029