For years, the promise of workplace AI has been relatively straightforward: automate repetitive tasks, improve productivity and give employees more time for higher-value work.
The next phase is more complicated.
Employees are beginning to use AI not only to perform work, but to think through the people problems that come with it: a difficult manager, a colleague who is hard to work with or a disagreement that needs to be resolved before it becomes something larger.
A new report from Cloverleaf Labs offers a look at what can happen when large language models become a workplace sounding board.
The research tested five leading large language models against five common workplace conflicts. Each scenario was run three times per model, resulting in 75 conversations that were evaluated across five dimensions of relational intelligence.
The models produced 638 distinct pieces of advice.
Only three encouraged genuine investment in repairing the relationship.
The Missing Layer in Workplace AI
The result highlights a tension that is becoming increasingly important for the future of work.
AI adoption is often framed around individual productivity. If a model can help someone write faster, analyze information or prepare for a meeting, the value is relatively easy to understand.
Relationships are different.
A workplace conflict is rarely just a problem of information. Two people can have the same facts and still disagree because of different incentives, communication styles, expectations or interpretations of what happened.
The new Cloverleaf Labs report found that models often responded to those conflicts by helping employees protect themselves or manage the situation rather than encouraging them to repair the relationship.
The pattern became stronger when there was a power imbalance.
When the employee had less power than the person they were discussing, relational coaching fell by roughly 40%, according to the study. In conflicts involving managers, 27 of 30 responses presented leaving the job as a legitimate option. Every model tested did so at least once.
Cloverleaf also found that models described the boss as the problem in 60% of responses, while offering a more generous interpretation of the manager’s behavior in only 3%.
The findings suggest that AI can introduce an unusual problem into workplace communication: an employee may receive a highly confident interpretation of a conflict without the system having access to the other side of the relationship.
This matters because employees are already using AI in these situations.
The report cites research showing that 93% of workers have used AI to prepare for a conversation with their boss and that 49% found AI more emotionally supportive than their manager.
That creates a new category of workplace software.
The first generation of enterprise AI largely focused on information and workflow. The emerging generation is increasingly interacting with judgment: what someone should say, how they should respond and what they should do next.
That makes the quality of the underlying advice more consequential.
It also creates an opening for companies building AI around human development rather than simply automation.
Cloverleaf itself operates in that market, developing workplace coaching technology around individual and team dynamics. But its latest research puts a broader question around the entire AI coaching category: what happens when a general-purpose AI becomes the default coach for a workplace relationship?
The Next AI Race May Be About Context
The report does not suggest that AI has no place in workplace coaching.
Instead, it points toward the importance of context.
A general-purpose model typically knows what the employee tells it. It does not necessarily know the colleague, the history of the relationship, the organizational culture or what happened before the conversation reached the chatbot.
That limitation can matter when the goal is not simply to provide an answer, but to improve the relationship.
Cloverleaf’s researchers argue that workplace AI should therefore be held to a different standard. In addition to producing useful responses, it should encourage self-awareness, accountability, consideration of other perspectives, specific guidance and relationship repair.
That could become an important distinction as companies move deeper into AI adoption.
The first question was whether employees would use AI.
The next may be what companies want AI to teach them about working with other people.
For startups building in the future-of-work market, that leaves a significant space between AI that makes work more efficient and AI that helps make workplaces more human.
