When clients of the cloud-based-software provider ServiceNow contact the company’s customer support center, 80% of the cases — in the form of calls and chat messages — are handled without any human intervention.
Instead, the company relies on analytical and generative artificial intelligence — in the form of AI agents — to address common customer questions.
Chris Bedi, ServiceNow’s chief customer officer and enterprise-AI advisor, said employees still handle one out of every five customer-support requests.
They’re getting new support from agentic AI, which can automate tasks such as drafting a response email to a customer. Workers remain in the loop for a final sign-off before any agentic-AI actions are executed. The combination of human workers and agentic AI shrank the amount of time it took to handle the more complex cases by 52% in a two-week period, ServiceNow said.
OpenAI’s cofounder Sam Altman and other leading technologists have said that 2025 will be the year that AI agents “join” the workforce.
In addition to ServiceNow, software developers such as Salesforce and SAP have rolled out their own agentic-AI platforms. These can perform workplace tasks such as processing customer invoices, providing customer support to clients, and drafting emails. The business software giant Intuit, which owns TurboTax and QuickBooks, began rolling out agentic-AI capabilities in December.
Humans mostly remain in the loop for now, but vendors anticipate this technology will become fully autonomous. Multiagent systems, where two or more AI agents collaborate to complete work, will proliferate.
“Agents are the next level of understanding around how you apply AI,” Jim Rowan, the head of AI at the consultancy Deloitte, said. “It can perform actions for you.”
In a recent Deloitte survey of 2,773 business leaders, 26% of respondents said their organizations were exploring autonomous agents to a “large or very large extent.”
For the first two years of the generative-AI boom — which kicked off after the debut of OpenAI’s ChatGPT in late 2022 — most businesses that adopted the technology scaled it to power chatbots and complete routine tasks like drafting meeting summaries. AI agents represent an evolution of generative-AI technology, built to complete tasks autonomously, though most are still monitored closely by workers.
Agentic AI “actually possesses some unique skills around reasoning, planning, and orchestration,” Bedi told Business Insider. “These agents can collaborate with each other and really start to deliver on the promise of work happening autonomously.”
Buzz for AI agents kicked into high gear after Salesforce debuted Agentforce in September to automate tasks in customer support, sales, and marketing. The company has said it will roll out 1 billion agents to customers by the end of this year. The company also reported that more than 340,000 of its customer support questions had been answered autonomously with Agentforce.
ServiceNow estimates that the company’s AI agents, already deployed in various parts of the business, such as customer service, human resources, and IT, are driving an estimated $325 million in annualized value by bolstering workplace productivity by 20%. ServiceNow says AI-agent-supported work saves 400,000 labor hours annually.
Still, technology companies are in the early stages of their agentic-AI development. Many are figuring out which processes they can fully automate with the technology. As a result, company leaders implementing agentic AI are training their workers to collaborate with — and provide feedback on — their new “coworkers.”
John Kucera, the senior vice president of product management at Salesforce, recommended that businesses be transparent about what work AI agents can handle and what will remain with workers. He added that businesses should be clear about what an AI agent actually is, saying that not all agentic systems are equal.
“There’s a lot of false agents out there,” Kucera said. “It’s only an agent when you’re taking a request and the agent is figuring out what to do and then what data to put in.”
While surveys frequently find that many workers worry that AI will replace them, technologists say AI agents won’t replace people but assume responsibility for mundane tasks.
“These agents are going to help me do my job, but at no point will they make me do something I’m not aware of,” said Walter Sun, the global head of AI at SAP, which sells software for financial, supply chain, and other business management needs. “The most important thing is that the employees are always in control.”
To ensure workers have a voice in how AI agents are developed, SAP encourages employees across its various business lines — including the travel- and expense-management provider Concur and SuccessFactors, which provides HR, payroll, and talent management software — to use an internal online form to reach out to the AI team and propose compelling agentic use cases.
At Intuit, the AI-powered financial assistant Intuit Assist can get businesses paid 45% faster by detecting past-due invoices and automatically drafting a personalized reminder note. After a business owner approves the note’s language and sends it out, they are paid, on average, five days sooner than with a human-only process, Intuit Assist said.
But before Inuit Assist takes action, humans have the final say. “What we’re trying to do is have the right human-automation interaction,” Ashok Srivastava, Intuit’s chief data officer, said.
Intuit has embraced a robust AI-training program, focused on responsible AI and what the technology can and cannot do, and built a “sandbox” called GenStudio that allows employees to interact with large language models in a secure environment. The company has also developed educational programs tailored to senior executives, directors, and engineers. “It’s very pervasive across the company,” Srivastava added.
Asana, which makes work-management software, launched AI agents in October, focusing on a few functions, including marketing, IT, HR, and research and development. Rather than track a specific number of actions that agentic AI takes over, Asana monitors the types of work that can be automated, eliminating the drudgery of busy work to allow employees to focus on more complex tasks.
The company is also keeping a close eye on which tasks AI systems get wrong compared with people. In cybersecurity, human errors tend to occur later in the day, when workers are tired after a long shift. AI doesn’t get tired, but it is susceptible to hallucinations — or when an AI model generates a response that is misleading or false information but nonetheless presents it as fact. For example, Asana’s AI agent might respond to certain questions by suggesting tasks that are, in reality, nonexistent to a particular workflow.
“The kinds of errors we see are different, so the way we fix them needs to be different,” Asana’s chief information security officer, Sean Cassidy, said. He said the company conducts automated tests to detect hallucinations and improve the product when they occur.
For AI agents to create a compelling return on investment for the companies that implement them, Deloitte’s Rowan said, workers should be tasked with frequently checking on simple automated tasks before any agentic AI actions are taken.
If businesses want to see big returns on their agentic AI investment, they need to place AI at the center of their work model, and then consider how humans will engage with the work, Rowan said. If not, “the savings really won’t be there,” he added.
ServiceNow’s Bedi said the success of AI agents depends on companies nailing three factors: New agentic capabilities should be developed for each department and its specific needs, unique training plans should be designed for every part of the business — like finance, marketing, and sales — and the value and return on of agentic workflows should be closely monitored.
“The companies that combine all three of those ingredients are going to have a competitive advantage,” Bedi said.
Read the original article on Business Insider
