The momentum of generative AI casts a shadow of uncertainty over the future of IT service desks

The rapid development of generative AI has sparked widespread discussion: as technology becomes increasingly capable in routine tasks, what role should humans play, or even whether human involvement is still needed. This issue is particularly prominent in entry-level positions in IT service desks.

Most analysts and technology leaders believe that generative AI will inevitably lead tojob disruptionacross the economy. However, when directly asked whether technology will replace workers, most people are cautious about this—some think it is possible, but not a wise move at present.

"It depends on the situation, but if you have a good knowledge base and are well-trained, it can indeed largely replace level-one service desk agents—which may be somewhat unsettling for some people," said Mark Tauschek, vice president of infrastructure and operations research at Info-Tech Research Group.

Level-one agents are typically the "front line" of the IT department, responsible for communicating with users and performing basic troubleshooting. This role is also an entry point for aspiring technical professionals. According to data analysis from Glassdoor (most recently updated in June), theaverage annual salary for level-one service desk employeesin the United States is approximately $48,000.

Although it is an entry-level position, level-one service desk employees have a significant impact on how other departments in the company view the IT department. Their efficiency in solving problems and the way they treat end users are thefoundation of employee experience。

However, experts are also thinking about the future role of these agents in the technology department, although many analysts believe that concerns about mass unemployment are exaggerated.

According to a survey of 405 U.S. director-level and above employees commissioned by Insight Enterprises and conducted by The Harris Poll in June, more than a quarter of professionals said thatworkforce displacement is one of their concerns。

Jeremy Rafuse, vice president and head of digital workplace at software development company GoTo, said that internally, concerns about job disruption among IT employees have intensified as teams seek to automate some tasks with generative AI.

"I think it's hard not to have that concern when you talk about automating certain workloads immediately," said Rafuse, who manages GoTo's IT department. "We are trying to promote this as an opportunity to learn new things, which may not only enhance your skills and allow you to do different work, but will also create jobs that do not currently exist."

GoTo is a sister company of LastPass (both now independently owned by Francisco Partners) and has automated routine tasks in its service desk for years. Recently, the IT team has invested time in learning generative AI and identifying low-risk use cases, Rafuse said.

"We don't want to just hide under a blanket," Rafuse said. "But the team is aware of it and is quite optimistic that this is an opportunity to learn new things."

In the service desk ecosystem, companies hope to use generative AI to analyze large amounts of data, identify trends related to satisfaction ratings, and pinpoint customer pain points.

"It would be very foolish if we didn't stay ahead in this area," Rafuse said, a view shared by most technology executives.

According to a survey of 900 respondents sponsored by Teradata and conducted by IDC in July, nearly 95% of technology executives said they feel pressure toadopt generative AI within the next six months to a year. More than half said the pressure is "high or significant."

Can ≠ Should: Calls for Cautious Advancement

Despite concerns about generative AI replacing workers, popular model makers have pushed back against the narrative that LLMs and generative AI capabilities should—or can—fully replace employees.

"Humans should always be involved in AI systems," Sandy Banerjee, head of GTM at Anthropic, said in an email. "Even if AI systems are answering questions or classifying tickets, there should be a quality assurance process where humans monitor the system and evaluate its outputs."

After all, AI is not without flaws. In research released by Anthropic, the company found that Claude models still make mistakes and fill knowledge gaps with fabricated content, emphasizing that models should not be used in high-risk situations or where incorrect answers could cause harm.

Researchers at Stanford University and the University of California, Berkeleyfound that models made by OpenAI do not necessarily improve over time. In some cases, performance and accuracy significantly declined, indicating the need for continuous monitoring.

Even so, models from generative AI startup Anthropic are available to enterprises through off-the-shelf methods and third-party services such as Slack. Asgenerative AI model providerscontinue to release updates in beta and treat tools as "work in progress," they are alsotargeting enterprise applications。

When OpenAI first launched ChatGPT plugins in March, it compared the code interpreter plugin in its blog to "a very enthusiastic junior programmer that works at the speed of your fingertips," but the tool can still generate incorrect information or "produce harmful instructions or biased content," as noted on the ChatGPT web page.

Despite warnings from vendors, the decision to replace workers with generative AI depends on costs and how organizations value specific roles.The National Eating Disorders Associationmade headlines for deciding to shut down its human-staffed national helpline and instead use a chatbot named Tessa. After usersshared experiencesthat Tessa suggested weight loss and dieting, thenonprofit organization removed the tool in June。

At GoTo, the company has set expectations through conversations and meetings about what generative AI technology can and cannot do. Rafuse emphasized that the technology should be used as a tool, but not fully relied upon because it can make mistakes.

"You know, sometimes you only get one chance, and if you tell someone the wrong information, they won't come back to you," Rafuse said. "First impressions are permanent."

Misinformation and bias are not the only risks. Before and during deployment, Melanie Siao-Si, vice president of international care and service at GoDaddy, said the team ensures communication of usage guidelines to prevent employees from exposing proprietary information.

The U.S. Federal Trade Commission (FTC)is currently investigating OpenAIto determine whether it has engaged in unfair or deceptive data security practices.

"Especially in care organizations, we are figuring things out as we go, so we are experimenting," Siao-Si told CIO Dive. "Obvious challenges include data security and privacy, as well as potential threat actors using the same technology to target our customer care organization."

Current State: Between Expectations and Reality

Eight months after ChatGPT's debut, most technology leaders describe their perception of generative AI technology as being between the "peak of inflated expectations" and the "trough of disillusionment" in Gartner's hype cycle.

The hype cycleidentifies five key phases: innovation trigger, peak of inflated expectations, trough of disillusionment, slope of enlightenment, and plateau of productivity. On Wednesday, Gartner placed generative AI at thepeak of inflated expectations。

on its 2023 emerging technology hype cycle. "Many technologies do follow this cycle, and I don't think generative AI will be any different," said Justin Falciola, senior vice president and chief technology and insights officer at Papa Johns. "The only question is that no one knows exactly which stage you are in right now."

The task for technology leaders is to navigate through the hype and find value. At GoDaddy, Siao-Si is leading AI experiments in the customer care team, and the company hopes to improve customer experience by providing multiple ways to contact and resolve questions or inquiries.

According to Siao-Si, on product pages, the team manages a prompt library to help customers plan their businesses in different ways. AI-driven bots also assist customers in refining the type of help they need before receiving guidance.

Enterprise software provider Atlassian announced in April that it updated the capabilities of its generative AI tool in Jira Service Management to support adjusting the tone of replies, as part of itsAtlassian Intelligence layerseries of updates.

"The reminder here is that the user is always in control," Sherif Mansour, head of AI products at Atlassian, told CIO Dive in an April demo. "They can dismiss and not accept suggestions, or they can accept, edit, and modify."

As enterprises prepare systems for generative AI implementation, leaders will have to addressknowledge gaps across the organizationand differences in user preferences. If companies fail to provide a seamless transition from chatbots to human touchpoints, it could also haveunintended impacts。

on customer experience. "If people don't like it, they will avoid it," said Chris Matchett, senior director analyst at Gartner. "That's why our advice to them is not to make it a 'gatekeeper' or 'greeter' that you must pass through to talk to a human."

Matchett said that generative AI-driven self-service channels should be implemented as an option for end users. "Let people choose it when they want it, rather than forcing it on them."