Artificial intelligence agents — tools designed to complete tasks with little or no human supervision — are dominating technology conversations, stealing at least some of the spotlight from generative AI chatbots like ChatGPT.

Software executives like Salesforce CEO Marc Benioff are lining up behind the technology, rolling out big plans for 2025 and beyond, as they imagine a future in which it’s trusted to perform a variety of key workplace tasks across industries.

Salesforce, for its part, is seeking to unleash one billion agents by the end of the year, Benioff said in an October announcement.

While the technology is generating a lot of buzz, some analysts say it may be a while before companies can begin to realize its enormous potential.

Agentic AI offers new opportunities for business leaders but it also poses a host of potential challenges, including the risk of an agent leaking sensitive data or acting based on a “hallucination” or error, according to analysts. This could hinder adoption in finance departments and other sensitive parts of a business such as human resources, at least for now.

“I think the finance function is going to need to take some time to get comfortable with agentic AI,” Mark McDonald, a senior director analyst at Gartner, said in an interview, adding that rapid adoption of the technology among finance teams is unlikely this year. 

Years not months  

The consulting group Forrester in a report late last year stated that AI agents, despite being a top emerging technology for 2024, were effectively not ready for prime time, asserting that it would be another two years before they have “any chance of meeting inflated automation hopes.”

In addition, some may be underestimating the complexity of running projects that involve multiple agents working together, according to Assaf Melochna, co-founder of Aquant, an AI company based in New York City.

“There’s a difference between building something for an experiment versus an enterprise-grade system that needs to be bullet proof to avoid error,” he said.

Still, despite agentic AI’s immediate challenges, analysts say it might also be risky for business leaders to completely tune out the noise around the technology, according to analysts.

“With agentic AI, the question is not if, but when,” Deloitte said in a 2024 year-end AI report released this week. “Although the technology is still in its early stages, it is evolving rapidly and will likely become increasingly capable over the next few years.” 

Over the past year, organizations have improved their preparedness for generative AI more broadly when it comes to areas such as tech infrastructure and strategy but have made seemingly little progress on other critical fronts, including risk, governance and talent, the report said.

Chatbots versus agents

The tech industry’s agentic AI push comes amid growing disillusionment over generative AI in the business world as many C-suite leaders grapple with the reality of cost and return-on-investment challenges, analysts say.

For organizations struggling to see the benefits of generative AI, Cole Stryker, a writer who contributes to IBM blogs, writing in an October piece said he sees agents as potentially “the key to finding tangible business value.” 

“It remains to be seen whether the vast sums of money currently being poured into a handful of huge LLMs [large language models] will be recouped in real-world use cases, but agentic AI represents a promising framework that brings LLMs into the real world, pointing the way to a more AI-powered future,” Stryker argued.

Chatbots “can interact intuitively with humans, synthesize complex information, and generate content,” according to a November article published by Deloitte’s Center for Technology Media and Telecommunications. “But they lack the degree of agency and autonomy that agentic AI promises … AI agents don’t just interact. They more effectively reason and act on behalf of the user.”

Microsoft 365 Copilot, a ChatGPT-like tool, can serve as a personal assistant to business professionals by “jumpstarting” creative projects, but an AI agent can go even further, acting as “a virtual project manager” or tackling complex assignments like reconciling financial statements to close books, Susanna Ray, a features writer who produces online content for Microsoft, wrote in a November blog post.

“An agent takes the power of generative AI a step further, because instead of just assisting you, agents can work alongside you or even on your behalf,” she said. “Agents can do a range of things, from responding to questions to more complicated or multistep assignments.”

AI agents can choose a course of action, design a plan, gather relevant data, and employ multiple software tools to complete each step, according to an SAP explainer.

“By learning from evolving information, AI agents improve over time — analyzing feedback, self-correcting errors, and solving emerging problems,” according to the article. “Multiple AI agents can also work together, even coordinating with humans to achieve different tasks.”

SAP supports the policy that a “human should always be in the loop” when agentic AI is used, Walter Sun, global head of AI at SAP, told CFO Dive. He said SAP’s AI agents provide recommendations to users, who decide whether or not to take action. 

While agentic AI models are designed to make decisions and carry out complex sequences of actions, they are not foolproof and can still make mistakes, just as humans do, according to Mark Purdy, managing director of Purdy & Associates, an advisory and thought leadership firm.

“The agentic AI prize could be great, with the promise of greater productivity, innovation, and insights for the human workforce,” he wrote in a December article published by the Harvard Business Review. “But so, too, are the risks: the potential for bias, mistakes, and inappropriate use.”

Organizations can begin preparing for agentic AI by developing a “strategic roadmap” and assessing which tasks and workflows are well-suited for the technology, according to Deloitte’s year-end AI report.

“[W]hile there are still many challenges to overcome — and technical complexities to sort out — now is the time to start preparing,” the report said.

AI’s ‘staying power’

More broadly, AI remains a top priority for C-suite leaders, despite some of the potential challenges and risks around adoption, KPMG found in a survey released this month.

The majority of corporate executives responding to the poll said they expect AI to “fundamentally” change the nature of their business within a year or two.

And nearly seven in 10 respondents said they plan to invest between $50 million to $250 million in generative AI over the next 12 months, up from 45% in first quarter 2024 poll results.

“The results show the staying power of the technology,” Todd Lohr, head of ecosystems at KPMG’s U.S. advisory division, said in an interview, adding that C-suite leaders have continued down the path of AI investment, by and large.

The coming year brings an opportunity to scale and advance AI capabilities across the enterprise, and a majority of organizations are looking to AI agents to help do so, KPMG said in a release announcing the study. Over half (51%) of organizations are exploring the use of AI agents and another 37% are piloting them, it found.  

But, once again, it will be up to businesses and their leaders to carefully weigh the risks and the benefits of various AI options.

CFOs and CIOs in particular are responsible for working together to ensure the company is aligned with the right projects, Melochna said.

“On the one hand, you have this pressure to invest to avoid falling behind competitors,” he said. “On the other hand, you need to be cautious to ensure you’re investing in the right thing.”