CIOs are bullish on AI agents. IT employees? Not so much.

Most CIOs and CTOs are bullish on agentic AI, believing the emerging technology will soon become essential to their enterprises, but lower-level IT pros who will be tasked with implementing agents have serious doubts.

While 53% of tech execs see AI agents being core to business operations in the next two years, only 29% of IT practitioners see that coming to pass, according to a new survey from uptime monitoring vendor PagerDuty.

Skepticism from IT practitioners is natural, given they will have to work out how to deploy and maintain AI agents, says Tim Armandpour, CTO at PagerDuty. While it’s the job of CIOs and CTOs to scout out new technologies, IT managers and staff are “closer to the action and closer to the madness of what it takes to operate and manage things at scale,” he says.

The survey, of IT professionals at the director level and above, suggests CIOs and CTOs will need to educate employees and promote agentic technology to them, some IT leaders say.

Truly autonomous agent technology is still in its infancy, with few organizations deploying sophisticated and fully featured agents, some experts say, but the divide between C-level IT leaders and their employees could create problems as adoption soars in coming years.

Widespread adoption on the way

Overall, 38% of respondents to the PagerDuty survey believe agents will become a core technology in the next two years, while another 50% see agents as a peripheral technology in the same timeframe. The PagerDuty survey aligns closely with a recent survey from Salesforce’s MuleSoft, which finds that 93% of enterprise IT leaders have deployed or plan to deploy AI agents within two years.

While C-level tech execs are enthusiastic, agent technology is probably more complex than many IT leaders now understand, with products currently sold as agents using “highly deterministic and prescribed” workflow paths, says Maya Mikhailov, CEO of Savvi AI, vendor of an AI deployment platform. True AI agents will have the freedom to decide on the processes they run and on tool usage and learn from these decisions, she adds.

Deployment takes a lot of work

Some of the resistance is likely coming from IT professionals who comprehend the work needed to deploy them, Mikhailov says. Other AI experts have warned organizations about building AI agents without outside help.

“Frontline AI practitioners have likely seen the amount of customization, quality assurance, and maintenance required to make a somewhat functional agentic workflow,” Mikhailov says. “Although the future state may involve the AI agent writing the code and connecting to systems by itself, it now consists of a lot of human labor and testing.”

Another reason for hesitation from IT practitioners is a lack of expertise in deploying and monitoring the technology, says PagerDuty’s Armandpour, who believes some natural skepticism will dissipate as IT workers get their hands dirty. There aren’t a lot of places to learn about agents now, other than hands-on experience, he adds.

“Many folks are going to be, in a way, almost forced to learn,” he says. “Once that starts to grow, it will be this grassroots type of adoption, and there’ll be enough CIOs, CTOs, and companies that would be willing to lean in because, honestly, you’re not going to have a choice.”

In addition to concerns about a shortage of skills and deployment headaches, many IT practitioners have seen other AI projects fail in recent years, adds Rahul Chahar, co-founder and CTO of Pull Logic, a provider of inventory management software. After IT professionals have witnessed several “overpromised” AI initiatives fall flat, they may be hesitant to commit to other large-scale rollouts, he adds.

“IT practitioners are cautious due to concerns around accuracy, transparency, security, and integration complexities,” says Chahar, echoing Mikhailov’s critiques. “Agentic AI systems often prove unpredictable, tough to troubleshoot, and challenging to mesh with older infrastructure, not to mention the compliance and security headaches they can create.”

Pull Logic ran into its own difficulties when trying to deploy agentic AI on behalf of a distributor selling products to building contractors. The customer wanted an AI-powered tool to recommend substitutes when products were not available, but the difference in product specification descriptions between manufacturers led to inaccuracies, Chahar says.

“During testing, the AI began hallucinating data due to inconsistencies in catalog structures,” he adds. “It started fabricating product numbers and features, which could have led to severe business consequences if left undetected.”

Start with a problem to solve

To sell agents to their staff, CIOs and CTOs must approach the new technology focused first on the problems they’re trying to solve, Chahar adds.

“CIOs and CTOs need to align agentic AI initiatives with their teams’ real challenges,” he says. “This means adopting a problem-first approach rather than chasing AI for the sake of it.”

IT leaders should integrate human expertise into the adoption process, while prioritizing transparency and investing in training, he adds. “Rather than pushing AI top-down, leaders should showcase tangible, incremental successes, establish clear accountability frameworks, and cultivate internal advocates to drive trust and widespread adoption,” Chahar says.

CIOs and CTOs should also deploy AI agents in phases, starting with small but high-impact projects that can deliver quick wins, adds Rob Kim, CTO at Presidio, an IT solutions provider. They should also welcome regular feedback from IT employees.

“Create regular forums for practitioners to share insights and challenges with leadership,” Kim says. “This ensures that leadership is conscious of ground-level concerns and can tackle them proactively.”

IT practitioners wary of AI agents should also remember the technology will improve as it matures, Kim says. As model reasoning improves, true multi-agent collaboration will allow agents to take more initiative to achieve the IT team’s desired results.

“This will be the worst version of agents and agentic AI that you will ever use ­— it only gets better,” he says. “Humans will then be able to orchestrate a larger volume of parallel workstreams and preside over the work of agents and bots.”