Navigating relationships can often have their difficulties, and the budding relationships between employees and their new AI coworkers are no exception. Employees aren’t exactly shunning their new coworkers, and in some cases may even be developing unhealthy relationships.

Looking at these relationship issues from different perspectives, three commentators suggest that AI coworkers require far different protocols than human workers, yet in some ways, still act as confidants.

“What actually changes when AI stops being a tool you use and starts being a co-worker?” was a question recently put forth by Ved Sen, head of innovation for TCS UK and Ireland. For starters, AI is nothing like a human coworker both in structure and capabilities.

The relationship between workers and AI coworkers may even be going beyond the bounds of professional working relationships. Employees are turning to AI “for personal support, including career advice and emotional validation – things that coworkers traditionally provide.”

That’s the word from Constance Noonan Hadley and Sarah Wright, writing in Harvard Business Review. More than half the 1,545 knowledge workers they studied, relatively far along the AI adoption curve, report they were “lonely at work,” and were relying on AI for social support. Such unsettling relationships "might erode company cultures and coworker cohesion in the future,” they warn,

The “co-worker as a system is a very different kind of being,” Sen advises. “It can be much more omniscient” when it comes to task performance. At the same time, AI coworkers have less flexibility in how they interact – requiring "clear ownership, a defined scope, an understood failure mode, and someone accountable for the outcome,” he observed. “Most enterprise AI deployments today have none of those things cleanly established.”

Then there’s the interface question. “If ambient AI handles much of the execution layer, the human role in that loop shifts – you’re no longer operating the system - click by click. You’re now instructing it. That’s a big change in the role and skills,” said Sen.

This is part of the decline of the traditional user interface. “Buttons, dashboards, approval queues: all of it was built on the assumption that a human would see the output and take the next step, based on constant operation and review,” Sen said. . "What replaces it is probably something more like a more sophisticated goal specification. You define the outcome. The system figures out the path. You need to check in for governance, not operation.”

To succeed in a workforce of AI coworkers, it’s important to be able to allocate and articulate the expectations of AI counterparts. "The skill that matters shifts from throughput to judgment: what should be done, for whom, under what constraints, and with what trade-offs,” Sen says.

Then there’s the design of work question, Sen speculates. “Who allocates? Who approves? Who overrides? Who sets the goal? Who catches the error the model made?” he asks.

People working closely with AI or AI agents even show a tendency to assign the machines human traits and even emotions, Hadley and Wright found. “They treated AI tools like humans. For example, 78% of participants used polite terms like ‘please’ and ‘thank you’ when interacting with them," they observe.

Even more telling is when asked to pick the analogy that best captured how they viewed AI at work, 28% of respondents to Hadley and Wright’s study chose humanlike terms such as “personal assistant,” “teammate,” and even “friend,” rather than technological ones such as “tool,” “platform,” or “engine."

Using AI more often and using voice mode instead of text prompts “intensified the tendency to personify AI,” they added. These AI assistants were credited with providing career advice which went beyond the scant advice provided by human managers in organizations, as indicated by 64% of respondents. Another 54% turned to their AI companions for advice on life skills. such as reducing stress, or “becoming a better listener, being more patient, and solving problems better.” Fifty percent even saw AI as a ”work friend" they enjoyed interacting with.

AI should be designed to promote human interaction, Hadley and Wright urge. Still, they advise against “overhumanizing AI – for instance, don’t assign AI agents names and personas to make them seem friendlier.”

Look for ways to introduce “positive friction into employees’ interactions with AI so that, when appropriate, the path of least resistance leads back to people instead of defaulting to AI," they also advised. This includes guidelines for when and how AI can be used to replace human interactions. “Employers need to determine when people should set aside AI in favor of human-to-human contact."