How to prepare your organisation for the rise of agentic AI

The emergence of agentic AI could transform organisational dynamics. It allows technology to take the lead in processes while humans provide essential oversight.

In brief

• Agentic AI goes beyond Generative AI. It represents a fundamental change in organisational roles and decision-making dynamics.
• Successful adoption hinges on balancing autonomy with oversight, driving cultural and mindset shifts, and embedding semantic business models alongside robust risk mitigation.
• Its true value emerges in complex, variable tasks where traditional automation tools fall short.

As organisations increasingly explore the capabilities of artificial intelligence (AI), agentic AI has emerged as a transformative force in the industry. Distinct from traditional AI and Generative AI (GenAI), agentic AI represents a shift in how we interact with technology, enabling systems to take on more autonomous roles in decision-making and process execution. It’s about going deeper with AI because the Future Won’t Wait. Tim Morthorst, Technology Consulting Director, Digital & Emerging Technologies at EY Ireland, examines the nuances of agentic AI, highlighting its applications, challenges, and the steps organisations can take to prepare for its future adoption.

Q. What are the primary types of agents in agentic AI, and how do they operate?

A. Agentic AI agents can be grouped into two broad categories: Assistant Agents and Automation Agents.

Assistant Agents are what most people envision when they think of GenAI or agentic AI. These are intelligent chatbots that engage in conversations with humans to receive short-term objectives. Based on these objectives, they apply reasoning and determine as well as execute appropriate actions by leveraging various systems and data sets at their disposal. For example, a travel agent chatbot might respond to a user who says they want to go somewhere sunny for a week. The agent applies reasoning by identifying sunny destinations, searches different travel agencies for packages, and then books the trip as well for the user.

In contrast, Automation Agents are designed to handle longer-term objectives and are not solely driven by human interaction. For instance, an Automation Agent might be tasked with continuously monitoring a backlog of work. In this case, it will only initiate a conversation with a human when it identifies a situation that requires human intervention, effectively instigating the dialogue rather than waiting for a human to prompt it.

Together, these two types of agents illustrate the diverse capabilities of agentic AI, enabling more efficient and proactive interactions in various operational contexts.

Q. What are the most promising real-world applications of agentic AI?

A. Currently, the most immediate applications of agentic AI are found in areas where organisations have struggled to address complexities using traditional automation tools like Robotic Process Automation (RPA) and intelligent automation. These applications can be defined as high-value, low-risk, and low-effort, and include:

Agentic AI delivers truly open-domain capabilities, with applications limited only by human imagination. Its greatest value emerges in areas where complexity and variability have long resisted traditional automation approaches.

Q. What challenges and risks are associated with the deployment of agentic AI?

A. Deploying agentic AI comes with several important challenges and risks that organisations must address before implementation.

One of the primary concerns involves data security as well as issues like GenAI hallucinations. These are risks frequently highlighted in media discussions about AI. While these risks are significant, effective mitigation strategies can be integrated during the solution design phase to address them.

Another challenge is the necessity of defining a clear semantic business model. While it may seem straightforward to provide an agent with a task and expect flawless execution, it is crucial to recognise that AI agents lack certain human attributes, such as cultural understanding, awareness of organisational goals, and familiarity with specific terminology and abbreviations. A semantic business model serves as a structured description of key business elements that provides agents with the tacit knowledge required for accurate execution.

Traditional automation is rule-based, whereas AI agents introduce a degree of autonomy. Granting full autonomy, however, poses unacceptable risks in most organisational contexts. Instead, controlled agency is essential to make decisions and take actions within defined boundaries that balance risk reduction with operational benefits. This can be managed through design and prompt experimentation in the semantic business model as well as dynamic response evaluation, and contextual grounding, but the complexity involved is often underestimated.

Lastly, and perhaps the most difficult challenge, is in shifting mindsets. Granting agency to AI agents, albeit in a controlled manner, can create discomfort as it may feel like relinquishing control. This psychological barrier can be difficult for organisations and their stakeholders to navigate, making it a critical consideration in the deployment of agentic AI.

Q. How can organisations position themselves for the future adoption of agentic AI?

A. As the conversation around agentic AI evolves, the most significant challenge organisations face in preparation is shifting their mindset. To fully leverage the potential of agentic AI, organisations need to move from people running processes with help from technology to technology running processes with people overseeing and supporting the systems. This represents a fundamental change in hierarchy — moving from humans as primary process drivers to humans as strategic overseers of technology-driven operations.

While organisations may initially recognise efficiency gains from this transition, it is crucial to understand that it will also disrupt existing business models. Organisations need to proactively prepare stakeholders for this shift as it will require a re-evaluation of roles, responsibilities, and expectations across the organisation. Organisations can unlock the full potential of agentic AI and thrive in an increasingly automated future by cultivating a culture that embraces this transformation.