AI can tackle benefits management. What about interpersonal issues in the workplace?

At some companies, workers might be split between those who go into the office and those who work remotely. But at Asymbl, a workforce orchestration company based in Texas, there are two camps: humans and AI agent employees—or, as the company calls the latter, “digital workers.” In fact, “digital workers” make up a majority of the company’s employees.

One of those AI agents, Polly People Ops, can answer questions and help employees make changes to their benefits, retirement plans, and PTO. The agent, accessible via Slack, was trained on Asymbl’s company policies and takes approximately 10 hours of work per week off the (human) HR team’s plate.

Polly is just one example of how agents are being incorporated into employee-facing roles. More than a third of companies are using AI in the realm of employee experience; a subset is using agents to help employees customize benefits and another for learning and development, according to the Society for Human Resource Management (SHRM). But adoption only goes so far: Although developers are rolling out agents for training and coaching employees, experts told Morning Brew they’re hesitant to allow AI into those types of endeavors.

Onboarding AI

Though Asymbl’s 200 or so digital workers aren’t human, employees are encouraged to treat them as such. Doing so, CEO Brandon Metcalf told Morning Brew, helps employees think of the agents as their contemporaries.

“We don’t call them agents, because we think of them more as employees,” he said. “How we talk about things, why we name things, why we call it ‘onboarding’ and not ‘set up,’ and ‘coaching’ versus ‘prompting’ or ‘tuning’—because our employees, our human employees, get it. And they start to treat [the agent] like a coworker.”

But unlike a human, Polly can only answer questions based on the information programmed into it; it can’t dig for beyond-the-book answers or interpret how HR policies should be enforced. If the agent is asked a question or prompted to perform a task beyond its scope, Polly directs employees to a human HR employee.

Relatively basic agents like Polly are becoming increasingly popular in the corporate world, SHRM Chief Knowledge Officer Alex Alonso told Morning Brew. Such AI applications can help employees compare benefits, take advantage of wellness perks, and set up parental leave.

“[Agent functionalities] help design that experience for the worker,” Alonso said. “That’s absolutely critical as we see people really trying to enhance their employer value proposition, because offering that flexibility and that customization for a personalized employee experience really makes a huge difference when you’re talking about impact overall.”

For Asymbl’s HR team, Polly makes an impact. Greg Symons, the company’s COO and founder, told Morning Brew that Polly was initially created when the company hired 100 people in 100 days. With the help of the agent, Asymbl’s HR team no longer answers basic people ops and onboarding questions. Now, they have more time for “meaningful opportunities” to “engage” with (human) employees, he said.

“Giving [the team] that time back allows for us to think about what other processes we have within HR right now that we can go back and look at strategically and say, does this process make sense? What stages in this process could be accomplished through AI, a digital worker, etc.?” Symons said. “It allows for them to continue to focus on moving the needle in terms of adopting and leaning into our own hybrid workforce methodology.”

Programming empathy

ServiceNow, a corporate AI software company, has also created HR-tailored agents that can be licensed for use by other businesses. Like Polly, their agents—which employees can communicate with on Slack, Microsoft Teams, or even via phone call—can help employees with tasks like requesting PTO or changing their name as it appears in company records. ServiceNow’s manager agent can also perform higher-level HR duties, like coaching an employee if they’re having issues with their manager or vice versa, or being the first point of contact for reports of workplace misconduct.

In such instances, John Phillips, ServiceNow’s GVP of product and GM of HR and employee experience, told Morning Brew that the agents can act as intermediaries between humans when they’re struggling to connect.

“Engagement is down and getting worse, and managers are stretched more and more thin. One of the highest complaints in turnover attrition is, ‘My manager lacks empathy.’ I would say that I can get a better empathetic response out of my AI than I can out of humans on a more regular basis right now,” Phillips said. “Most managers aren’t trained in empathy, and they aren’t trained in compassionate interaction. We’re at a tipping point where AI is going to help the manager be more empathetic.”

It’s all a simulation

While ServiceNow creates agents that it sees as human-like companions, workplace experts who spoke with Morning Brew agreed that empathy exhibited by agents isn’t as valuable as that from humans. Ann Skeet, the senior director of leadership ethics at Santa Clarita University’s Markkula Center for Applied Ethics, told Morning Brew that’s because the empathy agents exhibit is performed.

“The value that I put on that data is very different when it’s coming from an AI than when it’s coming from you. And the validation I might feel by telling somebody my story or asking my question directly to a human isn’t there when you’re interacting with an AI,” Skeet said. “The AI might simulate that kind of reaction, it might make you feel like you’re being listened to and heard in a certain way. It might try and use some words that express empathy or sympathy, but it’s still a machine.”

And because agents can’t react with facial expressions, Skeet said that results in a “flattening” of the social data communicated to the human. That’s why she suggests that companies “be thoughtful” about how they deploy AI to avoid eroding “relationships and human connection to each other.”

“The ability to collaborate, to improvise, to work together—all of those things that are uniquely human and also involve some human judgment, where you need to have human wisdom come forward, not just an algorithm that’s predicting what’s likely to come next,” she said. “The fundamental principle is keeping the choice about whether or not to use AI grounded in human dignity.”

Current users of ServiceNow’s agents seem to have taken similar principles to heart.

Georgia-based Wellstar Health System uses WALi, an agent that helps clinicians and providers with HR matters, among others, through Teams and the company’s intranet. Dianne Kokotoff, Wellstar’s executive director for enterprise solutions and automation, told Morning Brew that the agent has produced a 300% return on investment, but that “for complex issues that need empathy, nuance, or decision-making about patient care or sensitive HR matters, [human] agents and HR case workers are still in the loop.”

“We also have to be thoughtful about where AI takes the lead and where human judgment remains essential,” she said in an email. “We don’t put AI in charge of conversations that require real judgment or human connection.”