The Hitchhikers Guide for Driving Value Realization with Agentic AI

Who could be the winners from the pivot toward Agentic AI, and what will drive value realization from investing in it? PAC discussed these questions with Boris Krumrey, who leads UiPath’s Innovation Labs, when we met to discuss their progress. PAC had followed UiPath pretty much from when they came out of their garage in Bucharest to becoming the market leader in RPA, one of the fastest-growing segments in enterprise software ever, culminating in their well-documented IPO.

Yet, UiPath’s corporate development has been chequered from the heights of the IPO. Those turbulences led to the return of its founder, Daniel Dines, as CEO, with the goal of regaining UiPath’s mojo. The main lever to find that mojo is to expand from the heritage in RPA to Agentic Automation and ultimately deliver Agent Orchestration. For UiPath, their agents and robots work across all applications, both new and legacy, thus eliminating vendor lock-in concerns for customers. In contrast to ISVs like Salesforce, UiPath argues that business processes don’t run on a single system. Therefore, as Daniel Dines put it in UiPath’s most recent earnings call, “we will continue to be the Switzerland of business applications and agents, providing equal access to third-party systems.” The second lever for differentiation is seen in governance. Thus, the evolution toward agentic orchestration is meant to lead to becoming the conductor or the governor across agents, robots, people, and models.

Regarding the technology foundation for this pivot, the expansion from Agentic AI to Agentic Orchestration is akin to extending RPA to Intelligent Automation. Thus, the focus is not on building or integrating specific agents but on managing and orchestrating the plethora of emerging AI approaches. The emphasis is on providing a controlled agency with auditability, governance, and a trust layer. By integrating a broad set of Generative and Agentic AI models, the UiPath platform is slated to provide collaboration between humans, agents, and an array of automation.

These discussions prompted PAC to expand its deliberations to learn from other secular shifts, such as automation and cloud when assessing the pivot toward Agentic AI. To accelerate the journey toward Agentic AI whilst realizing value from those investments, organizations need to focus on two issues. First, those discussions should not be about whether OpenAI’s Operator will replace RPA but about aligning AI and automation to business objectives. To accelerate the transformation journey, organizations should demonstrate organizational and personal benefits. In another way, the more important thing is to articulate the “why” of the Golden Circle model. What is your purpose, your motivation, and what do you believe in? Therefore, organizations should first develop a nuanced understanding of Agentic AI and what it is and what it isn’t. PAC will cover those levers in forthcoming research notes. Second, they should define the business objectives and outcomes that will lead to a reimagination of business and operating models.  We can learn this from other secular technology shifts in automation and the cloud: The biggest challenge in capturing value from cloud-native transformation investments lies in aligning technology and business priorities to develop a coherent strategy.

 Therefore, PAC has developed the hitchhiker’s guide for driving value realization with Agentic AI, as depicted in Figure 1. It is meant to help organizations pivot their assessment of Agentic AI toward business outcomes without getting lost in trying to understand the plethora of capabilities of specific models. Rather, it depicts critical areas outlining contingencies such as data integration and suggesting a more nuanced comprehension of agency and autonomy inherent to Agentic AI. Any assessment should focus on how the evolution of AI can impact and accelerate business models and, consequently, operating models. However, it is here where the technology providers have done a modest job as they tend to focus on technology capabilities. We urgently need customer journeys that highlight a successful reimagination of business and operating models by embracing the evolution of AI. Let’s touch upon those critical areas in more detail.

Figure 1: The Hitchhiker’s Guide for Driving Value Realization with Agentic AI

 Source: PAC 2025

Data integration depth: As cliched as the statement that any AI project will only be as good as its data quality is, the ability to integrate unstructured data into workflows is central to the progress with Agentic AI. And we should have learned from adapting to regulations like GDPR that the demands on data management are exploding. The complexity of achieving compliance is immense, with many national AI regulations mushrooming. The ultimate goal is to achieve a real-time data flow that can be integrated into workflows. Yet, especially where organizations are being held back by technology and process debt, this goal is often all too difficult to accomplish.

Controlled agency: Building on the implications of data management, having trust layers that include the monitoring of models and the orchestration of agents in order to scale deployments with probabilistic decision-making. However, those layers must also safeguard the outcomes delivered, not least where LLMs and agents are being integrated. Fundamentally, organizations must design a control environment that manages Digital Labor. That means a delivery strategy where humans, bots, and agents co-exist and collaborate. What was largely lip service in the early RPA days is now starting to become a reality but also a challenge.

Decision-making complexity: One key learning from the automation discussions is to be clear about the context and goals of automation. The agency inherent in Agentic AI to independently make decisions and execute actions must go beyond tasks if we are to progress toward end-to-end automation. Therefore, we must cut through the noise of Agentic AI and be clear on the scope of the deployments. Only if we focus on end-to-end processes can the disruptive nature of Agentic AI be realized.

Autonomy spectrum: Probably the least understood area is the autonomy with which agents can manage workflows. We are just scratching the surface of non-deterministic automation as organizations look to leverage probabilistic approaches such as LLMs. However, there is not yet enough progress in reasoning and planning to speak of goal-driven automation. To progress in that direction, we must better understand known process steps and be capable of automating unknown scenarios. Technologies evolving in that direction will be closer to Google DeepMind than any agentic models.

The innovation cycles that AI transforms are breathtaking. But for those innovations to be implemented, providers must support their clients in overcoming technology and process debt and drive deep organizational and cultural change. Agentic AI will invariably lead to a new quality of Digital Labor. But this disruptive change must be managed and governed. These are the discussions that PAC is aiming to lead.