Generative AI chatbots seem to offer an infinite array of capabilities. OpenAI’s ChatGPT invites users to “Ask anything,” while Anthropic’s competing Claude model asks, “How can Claude help you today?” They can do anything from analysing spreadsheets to drafting poetry.
But every so often, many models must admit: “I can’t do that.”
Call a restaurant to book a business dinner? No can do.
Scout sales prospects and send pitches? Not with your run-of-the-mill chatbot.
Each of these examples shows a limitation many generative AI products have: While they offer advice, analysis, and content, they can’t take much action. But a new wave of AI products promises to change that.
“My view is that it’s actually turning into some real practical products. It’s the step beyond reactive and predictive AI — it’s about being able to make autonomous decisions and for that AI to take action,” said Sarah Ghosh, FCMA, CGMA, a finance leader in the UK’s civil service and a former CIMA president.
Known as agentic AI, these emerging tools are empowered to attempt to reason through complex issues, interact with outside systems, and take action on behalf of the user, with significant implications for the finance function.
What does agentic AI do differently?
Agentic AI represents an evolution from earlier tech solutions. For example, robotic process automation (RPA) is heavily rules-based and may struggle to adapt to changing circumstances — while agentic AI promises to be far more capable of setting its own course and navigating challenges.
“Agentic AI, that’s really the next step,” said Richard Allmendinger, Ph.D., Professor of Applied AI at the UK’s University of Manchester (UoM) and co-founder of VeriBee, an AI cybersecurity spinout of UoM. “These systems, agents — they plan, they execute, and all that with limited to no supervision.”
Tech company Nvidia, which sells AI services and platforms and whose processors power most popular AI models, defines agentic AI as incorporating four key steps:
- Perceiving data from different sources, whether it’s sensors or databases;
- Reasoning by using a large language model to parse requests and come up with a response;
- Acting by executing tasks that use APIs (application programming interfaces) to integrate external tools and software; and
- Learning from the results of its actions and continuously improving.
But agentic technology is in its infancy, and tech companies have only recently begun to embrace the “agentic” label.
“Is agentic AI just hype? It’s brand new. No one quite knows what this is capable of right now,” said US-based Greg Wallig, founder of the AI-centric management consultancy Agentic Advisors and chief strategy officer for software development company THEIA Analytics Group.
One early application is in software development, with Cognition AI’s Devin promoted as the “first AI software engineer”, capable of planning and executing “complex engineering tasks requiring thousands of decisions” — allowing engineering teams to work towards bigger goals, according to the company.
The idea is also catching on with much larger companies like the Ford Motor Company and Siemens. By stringing together AI agents across multiple phases of product design and testing, Ford’s AI agents can turn a sketch into a 3D model and then predict the results of physics tests on it in dramatically less time than current processes, The Wall Street Journal reported. On the other hand, Siemens is looking to create AI agent systems that can independently execute complete industrial workflows.
Google and Microsoft both have debuted platforms to help businesses build AI agents. And, by some definitions, OpenAI is also adding agentic capabilities to its models, with its Deep Research tool attempting to reason through complex research questions.
What is the potential for finance?
Agentic AI applications for finance will likely build on ways that AI is already used. Potential examples, which are in various stages of development, are:
- Relationship management: AI agents could help to manage personal and professional relationships, keeping track of email conversations, reading an acquaintance’s social media posts, booking appointments, and preparing notes, Wallig suggested. “The technology is acting on your behalf,” he said.
- Dynamic pricing and buying: Algorithms already have the power to set prices and buy stocks, turning an algorithm’s suggestions into action. AI also is being built into platforms like LevaData and Fairmarkit, which help companies analyse their supply chains. Eventually, agentic AI could take a more direct role in negotiating with suppliers, placing orders, and more.
- Fraud prevention: AI already tracks and flags suspicious transactions based on large amounts of data. But in the future, agentic AI could do more contextual analysis and move more directly to block activity. “It’s not just flagging expenses but disputing them, maybe,” Allmendinger suggested.
- General automation: While RPA works well for rote and unchanging tasks, agentic AI could be more flexible, using contextual information to take the right action even for tasks that may be somewhat variable. AI already allows automation routines to work more easily with other systems, trigger decisions, and more, Ghosh said. “I’m seeing things shifting in terms of robotic process automation, with intelligence built into that,” she said.
- Tax and regulatory: AI agents could take a greater role in tax planning and regulatory compliance. “As new rules come out, you can imagine agents going through the system you have and making sure you meet the regulatory requirements,” Allmendinger said.
Agents’ roles could fall into several categories, including full automation of tasks; collaboration with humans; and in-depth data analysis, according to Boston Consulting Group (BCG). It predicts the market for AI agents will grow from $12 billion in 2026 to $52 billion in 2030 — a compound annual growth rate of 44% over that period.
What does agentic AI mean for finance now?
Agentic AI remains in a “trial phase”, Allmendinger said. Much of the technology already exists — combining existing advances in language processing, predictive analysis, reasoning, and more.
“The building blocks of agentic AI — they exist, and they are quite good,” he said. The question is “how do these building blocks interact with each other in a responsible manner?” he said.
It’s too early to say just how much of an impact the technology will have — and there’s no need for most organisations to rush into a deployment, he advised.
“There still aren’t many use cases,” he added. “People are using it more and more … but not everyone needs to be an early adopter.”
But Ghosh believes organisations need to be adopting the technology now. AI is already having “significant impact in organisations that are deploying it. It is reshaping the finance function and roles within it,” she said.
“There is a need for organisations to start to explore [agentic AI] as a mechanism to increase productivity and efficiency. It’s going be a critical part of how finance functions operate in the future,” she said.
Still, this latest tech wave adds to the sense that technology will take on more and more “human” roles in the workplace — adding to the automation trend that has already reshaped finance.
Riyadh, Saudi Arabia-based Rashid Khan, ACMA, CGMA, is a financial reporting lead for Mozn, a company focused on risk, compliance, and Arabic generative AI products. For now, much of Khan’s work remains manual, from transaction processing to analytics and balance sheet analysis. But he expects that to change — resulting in a shift in his role.
“I think agentic AI would do more of this on its own and give you the final results, where you as a decision-maker, you’ll have the time to make the right decision,” he said.
Seeing the technology emerge, Khan has changed his career priorities, focusing more on learning how to use AI to complement strategic decision-making.
“Instead of consuming myself in learning how to work on Excel, this is all going to be done in the foreseeable future by AI,” he said. “So how can I use this data? How can I make better decisions in terms of the business?”
Using and managing AI in that way will be a growing necessity for employees, according to BCG. People may even supervise teams of virtual AI agents, and they’ll need new management and technology training to do so, BCG suggested.
What’s the limit?
Wallig thinks this technology will change companies large and small, which is why his Agentic Advisors consultancy aims to make AI tools available to clients.
“The imagination is probably our only limit,” he said. But like others, he cautioned that giving AI agents autonomy will only amplify the potential that AI decisions will be tainted by mistakes, hallucinations, and biases.
Given known risks across generative AI, the issue is “really around control”, Ghosh said. “How do we, in the agentic AI landscape, ensure that we can monitor and override decisions made, and where does the accountability sit for making those kinds of decisions?”
These concerns underscore the potential for fraud and the need for security. “How easy is it to manipulate some of these [agentic AI] solutions?” Ghosh asked. “It’s building on the risk that we’d already identified through gen AI and just framing it around that autonomous decision-making. Because we do need to make sure we have that oversight and assurance as finance professionals, that we can intervene at the right point.”
Ghosh anticipates that for finance, agentic AI will first be used in “ring-fenced” and isolated tasks, such as the preparation of accounts.
“That feels like it’s a good start and quite safe, but it needs that human overlay to ensure we’re all complying with all the regulatory,” she said. “[Tasks] that involve payment of money out are riskier.”
Others wonder just how far it will go. California-based Leighton J. Smith, ACMA, CGMA, works as a fractional CFO and consultant for startups. He said, “[What] kind of control will you relinquish? What AI gives you is speed — but I believe decision-making, and interpreting insights within the broader business context and catering to typically nuanced C-suite’s personal requirements, still demand human judgement.
“AI can assist with the heavy lifting, but it takes experience to know what to ask, how to verify, and how to [re]act.”
For his part, Wallig believes the advance of AI will only add to the value of accountants and other finance professionals.
With more data being placed into “black box” technology, there will be a greater need for people who truly understand the underlying financial systems, he said.
“But you’ve got to understand how it works,” he said. “Really get smart on this technology. You will find yourself very valuable [in] this future world.”
— Andrew Kenney is a freelance writer based in the US. To comment on this article or to suggest an idea for another article, contact Oliver Rowe at Oliver.Rowe@aicpa-cima.com.
LEARNING RESOURCES
Designed for the busy finance leader, this bundle provides you with a baseline understanding of digital transformation and the technologies that are changing the finance function.
COURSE BUNDLE
Ethics in the World of AI: An Accountant’s Guide to Managing the Risks
This course discusses the current uses and risks of AI in business and provides practical suggestions to address these risks effectively.
COURSE
MEMBER RESOURCES
Articles
“AI Concerns Are More Complex Than Technology Replacing People”, FM magazine, 16 April 2025
“Finance Chiefs Look to Digitalisation to Cut Costs”, FM magazine, 9 April 2025
“The ROI of Generative AI: It’s Growing Rapidly, CFOs Say”, Journal of Accountancy, 21 March 2025
“Global Handbook on AI Safety Released”, FM magazine, 31 January 2025
Podcast episodes
“Big Data & AI Technologies: Fundamentals and Applications”, Reshaping Finance, 8 April 2025
“AI’s Future: Figuring Out What It Means for Finance Teams”, FM magazine, 22 January 2025
