Agentic AI: Key Points CFOs Need to Know

The following is a guest article by David Hickey, leader of Baker Tilly's intelligent automation practice, representing only the author's personal views.

If 2024 was the year of generative AI, then 2025 will be the year of Agentic AI. If you haven't yet learned about Agentic AI, that's not surprising. Think of Agentic AI as enhanced artificial intelligence, or AI version 3.0. You may have also recently heard the term "agent" in relation to AI. This description is quite fitting—as a super-intelligent technology, Agentic AI simulates human thinking during analysis and processing. This technology relies on large language models, natural language models, and specialized software used to write and train machine learning algorithms.

Differences between Agentic AI, RPA, and Generative AI

Agentic AI (often called agents) can make autonomous decisions and act toward complex goals with minimal human supervision, combining technologies such as large language models and machine learning. In contrast, Robotic Process Automation (RPA) uses software robots to automate repetitive, rule-based tasks, requiring human intervention for exceptions. Generative AI, on the other hand, creates new content such as text, images, or code based on user prompts, using deep learning to identify data patterns. In short, Agentic AI is highly autonomous and decision-centric, RPA is task-oriented, and Generative AI excels at generating creative content based on user input.

Agentic AI and RPA digital workers can work together to enhance the efficiency and decision-making capabilities of business processes. Agentic AI can handle complex decision-making tasks and adapt to dynamic environments, while RPA digital workers excel at automating repetitive, rule-based tasks. For example, Agentic AI can analyze data to identify patterns and make strategic decisions, then delegate specific tasks to RPA digital workers for execution. This collaboration enables businesses to leverage the strengths of both technologies, achieving higher levels of automation and smarter, more agile operations.

Practical applications of Agentic AI

Imagine the rumble strips alongside roads, which can help understand how to train agents. Humans set operational "guardrails" for agents, limiting the types of activities they can engage in. When an agent hits a guardrail, it steps back and informs the user that it cannot proceed, while reminding human colleagues that assistance is needed. As humans and agents collaborate, the guardrails expand as the agent learns.

Agents elevate customer interactions to an entirely new level. When combined with Agentic AI, robots can make decisions beyond preset matrices or flowcharts. Due to their proactive nature, agents can "think," reason, and adapt to dynamic environments without human instructions. In fact, agents improve their own thinking processes with each problem-solving iteration.

More interestingly, Agentic AI can help AI systems set goals, making them work smarter and more independently. Through continuous learning and process optimization, agents can adopt an organization's values, brand, and perspective.

Agentic AI is spreading across industries, with early adopters including well-known companies such as IBM, Apple, Siemens, FedEx, Duke Energy, UPS, Tesla, Goldman Sachs, and PayPal. The technology has been applied in areas such as customer service, medical diagnosis and personalized treatment plans, business process automation, manufacturing supply chain optimization, finance and algorithmic trading, smart grid and energy management, autonomous driving, climate modeling, and environmental protection.

What does Agentic AI mean for CFOs?

Agentic AI incorporates human-like decision-making into existing AI platforms, elevating efficiency to new heights and freeing employees from monotonous, repetitive tasks to focus on more important work. CFOs can rest assured that Agentic AI, like other forms of AI, is scalable and flexible. For example, organizations can embed it into customer-facing applications to provide highly customized experiences or advanced help desks; it can also be deployed in back-office operations. In many cases, mature enterprises integrate Agentic AI into both customer interfaces and internal operations.

Because agents continuously learn, self-improve, and build on past "experiences," they can handle complex, dynamic, and ever-changing scenarios. Their nature helps solve evolving real-world problems.

Will agents replace humans or lead to workforce displacement?

Similar to earlier versions of AI, we expect AI agents will not eliminate human contact. In certain complex cases, agent decisions require human oversight and review. What is known is that Agentic AI can free up employee time, allowing them to engage in higher-level, more strategic, and meaningful work. When deploying an Agentic AI strategy, business leaders also need to pay attention to ethical issues beyond employment displacement, including risks of misuse, unintended consequences, data bias, transparency, and data governance.

Like other emerging technologies, Agentic AI requires thoughtful, strategic implementation. This means starting with process identification, determining which specific processes or functions are suitable for Agentic AI. Business leaders also need to assess organizational value and impact, and establish evaluation and measurement mechanisms to ensure the technology delivers clear benefits. Companies should also pay attention to team composition, bringing in external experts when necessary to ensure successful implementation. Beyond technical feasibility, factors such as data security also need to be considered.

For now, CFOs and other business leaders need to understand the concept of "agents" and maintain an open mindset, thinking about how this powerful technology can best serve organizational needs. Whether ready or not, the new era of Agentic AI has arrived.