Less than three years after the introduction of ChatGPT, the next generation of artificial intelligence (AI) is coming, and it promises to change the way accountants do their jobs.
Agentic AI, as indicated by its name, is an AI that has agency. It has the ability to act autonomously and make decisions. It can set complex goals, plan the steps to achieve those goals, and adapt to changing circumstances with little to no human help.
Where generative AI tools like ChatGPT require user-provided, step-by-step prompts, agentic AI uses “sophisticated reasoning and iterative planning to autonomously solve complex, multistep problems,” said Randy Johnston, CEO and co-founder of Network Management Group Inc. (NMGI).
Agentic AI is on the verge of being incorporated into many AI tools and is expected to be widely used by year’s end, said Donny Shimamoto, CPA/CITP, CGMA, founder and managing director of IntrapriseTechKnowlogies LLC.
“It will free accountants up more for the analysis and strategy to help people make decisions,” Shimamoto said. “It will make our jobs easier and let us really do what we’re really designed to do and trained to do.”
He likened it to having a team of Ph.D.-level researchers at your beck and call who can work at lightning speed to think through complex projects.
Johnston, who will team with Brian Tankersley, CPA/CITP, CGMA, of K2 Enterprises to deliver the Technology Update session June 10 at the AICPA & CIMA ENGAGE 25 conference, foresees agentic AI agents being used for customer service and sales support, supply chain reconfiguration, and managing complex IT projects. After reviewing a few accounting-related agents as of this writing, he predicts that in the next one to three years, agentic AI will be able to:
- Automate key elements of client advisory services, including bookkeeping and report generation.
- Empower real-time auditing with enhanced detection of anomalies that might indicate fraud.
- Assist in directed research with superior results compared to the general large language model (LLM) AI tools like ChatGPT.
HOW AGENTIC AI WORKS
Generative and agentic AI use algorithms that predict which word is next and simulate certain aspects of human learning and decision-making. Both are trained on large datasets and respond to natural language requests, generating content that includes text and images. Agentic AI uses some additional programming to learn and adapt more quickly than generative AI and make decisions based on context, which generative AI cannot do (see the chart, “Key Traits of Agentic AI vs. Generative AI,” below.)
Generative AI responds to prompts such as “summarize this meeting and call out action items from the transcript” or “convert this report into a presentation with 10 slides.” Agentic AI can autonomously reason and take on multiple-step queries to seek out the most common or predictable answers and build upon that, according to Johnston.
In the case of a client presentation, for example, software using agentic AI could be asked to research recent tax law changes and then synthesize that insight into a presentation that can be shared at client meetings. Or it could be asked to produce a year’s worth of marketing materials, which would involve determining the audience and goals for the firm by analyzing strategic documents and prior marketing efforts and then creating content and scheduling its distribution on various channels. Agentic AI will also be able to manage customer or client interactions on social media platforms, websites, and other customer-facing platforms.
Agentic AI solves problems via a four-step process:
- Perceive: An agentic AI agent can input information from text, audio, cameras, and sensors to receive project requests from human users and also to gather data from various sources.
- Reason: The agent draws upon its LLM technology to process and understand the data it has gathered and to set goals for achieving the requested project.
- Act: The agent can then operate on its own to complete the tasks required to fulfill the user’s request.
- Learn: The agent uses machine learning to continuously improve and learn through feedback.
An IBM report comparing Agentic AI vs. Generative AI asserts that agentic AI’s “unique ability to learn and operate on its own make it a promising technology for organizations seeking to streamline workflows.”
Like generative AI, agentic AI agents improve the more they are used, said Marc Staut, chief innovation and technology officer with Boomer Consulting. “Training will make them demonstrably more effective and impactful.”
SUBTLE INCORPORATION
Agentic AI’s entry into our everyday life won’t necessarily come with signs popping up to announce, “You’re now using agentic AI!”
Instead, expect incorporation resembling how search engines such as Google adopted AI platforms. A user wasn’t necessarily told what powered their search for the best tacos in the area, they just received a curated list within moments.
For those in accounting and finance management, agentic AI will likely be phased into existing software, and users may notice more user-friendly interfaces that can produce more sophisticated results, Shimamoto said.
Also, agentic AI is being phased in as access is increasing to do-it-yourself (DIY) automation tools, said Byron Patrick, CPA/CITP, the CEO of VERIFYiQ, an automated bookkeeping platform, and the co-founder of TB Academy, an AI education company for accountants. The combination of agentic AI and DIY automation sets the stage for small firms and finance teams to improve processes by easily building automations and incorporating agentic AI in their operations.
To determine which existing processes could benefit from agentic AI and which are better left alone for now, Staut suggested accountants and finance professionals get familiar with the new technology and how it works.
It’s important to remain strategic about what software and vendors you go with to ensure it’s a smart move for your firm and company and not just go on a technology shopping spree and buy everything out there with agentic AI, he said.
“It may be a challenge to figure out the best agents to invest in, not just lean into having the most,” Staut said.
CAPACITY FOR CHANGE
In essence, agentic AI will be able to accomplish what robotic process automation (RPA) was promised to do but often missed the mark on because of the inflexibility and complexity that popped up when it was used, Patrick said.
Think of RPA and generative AI as individual musicians and agentic AI as an orchestra — capable of having multiple agents on their own decision-making missions and then synthesizing all those individual pieces together.
“Even a simple deployment that runs continuously and in parallel at scale can produce exponential improvements,” Patrick said.
In a profession that has long struggled to train enough CPAs to meet the demand that’s out there, agentic AI could make inroads in the staffing crisis by taking over the tasks that have added hours to the workday and kept firms and finance teams from taking on more work.
“These tools are going to increase our productivity and our value and our output,” Patrick said. “And possibly even give us time back to go on vacation with our families while the bots are doing their work.”
It’s important to remember that AI, whether generative or agentic, can’t do the creative thinking that humans do. At the end of the day, AI is still about following processes and looking for the most common or predictable answers — which aren’t always the right ones. Human oversight is still needed, and more strategic decisions and advice should not be handed off to the bots.
“Human creativity looks outside the box,” Shimamoto said. “AI generally looks inside the box.”
CHALLENGES AHEAD
Behavioral shifts will be needed for people to harness agentic AI’s potential, as well as a willingness to see what the technology can do, Patrick said.
Another hurdle will be for people to understand and accept the results delivered by agentic AI. Generative AI is known to make things up, also known as “hallucinations.” Agentic AI is less prone to those, but they are still possible. In addition, accountants likely won’t be privy to the processes agentic AI uses to analyze, reason, and create things on its own. This lack of transparency breeds doubt, and the accounting profession already tends to be skeptical.
“If things are happening in a black box, it’s hard to trust that output,” said Patrick, who expects it will take time for accountants to develop a better understanding of how AI can help and where caution is needed.
It’s important to determine where a human needs to be involved in the process. This includes determining critical decision points and thinking about processes that might be difficult to undo or might cause immediate issues if done incorrectly.
It will be even more important for accounting firms and companies to conduct due diligence when it comes to vendors of AI products, Shimamoto said. Organizations should know where data is stored, who has access to it, and whether all privacy regulations are being followed.
In addition, the following questions should be addressed: How functional is the product? How accurate is it? Is it fit for purpose? If you are not using it for the exact scenarios it was trained for, its output may be inaccurate.
Patrick suggests that organizations lean on their existing processes to vet vendors. CPAs should ask for a System and Organization Controls (SOC) 1 report if the agentic AI will be used in the financial reporting process and a SOC 2 report if the agentic AI will have access to personally identifiable information. It is critical to evaluate the SOC 1 or SOC 2 report and make sure it provides the right information to evaluate the AI system’s functionality and security and how it addresses concerns about data privacy or confidentiality.
Other assurances to look for are proof of compliance with the California Consumer Privacy Act (CCPA) and the European Union’s General Data Privacy Regulation (GDPR) requirements.
Agentic AI is not a passing phase. CPAs in all areas of accounting will need to consider how much they plan on using it and should plan on being ready to rely on it in the next couple of years, Patrick said.
“Get curious,” he added.
About the author
Sarah Ovaska is a freelance writer based in North Carolina. To comment on this article or to suggest an idea for another article, contact Jeff Drew at Jeff.Drew@aicpa-cima.com.
LEARNING RESOURCES
AI Strategy & Governance
Learn what organizations should consider when implementing AI solutions. Attend the conference on June 16 at the Gaylord National in National Harbor, Md.
CONFERENCE
AI-Powered Excel: Leveraging AI and ChatGPT for Supercharged Productivity
Learn how to streamline your Excel work, research, and documentation, saving you time and effort in your day-to-day tasks.
July 22, 1-2 p.m. ET and July 28, 10-11 a.m. ET
WEBCAST
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.
CPE SELF-STUDY
Navigating the AI Landscape: An Introduction to Generative AI and ChatGPT for Accounting Professionals
Learn how to use ChatGPT to streamline your workload, save time, and boost efficiency.
July 8, 9-10:20 a.m. ET
WEBCAST
For more information or to make a purchase, go to aicpa-cima.com/cpe-learning or call 888-777-7077.
AICPA & CIMA MEMBER RESOURCES
Articles
