Artificial intelligence has been gradually taking over tasks once thought to require human judgment, such as data analysis, forecasting, pattern recognition, and even creating original written content. Now, we stand at a new frontier that cuts closer to the bone of what many consider uniquely human.
“Soft skills” such as empathy, leadership, and communication are often framed as intuitive or “natural.” But in reality, they’re some of the most complex abilities humans develop.
Developers are beginning to build AI systems aimed at honing these abilities. Some are positioned as leadership coaches, while others stop short of advice and offer real-time feedback on tone, confidence, and emotional framing instead.
Professionals are increasingly experimenting with AI tools as a source of support during moments of stress, conflict, or self-reflection. What remains far less clear is whether those simulations can actually teach it.
To explore that question, we spoke with two experts who view the problem from different angles. Together, their perspectives reveal both the promise and limits of using AI to develop skills rooted in human perception, motivation, and ethics.
Why soft skills are harder than they look
Dr. Art Markman, a cognitive scientist and professor of psychology at the University of Texas at Austin, says that part of the reason soft skills are so difficult to cultivate is that most people never receive formal education about how human brains actually work.
“Our modern science curriculum is primarily biology, chemistry, and physics,” he said. “But it omits psychology, a rigorous scientific field in its own right.” As a result, he adds, people “don’t understand themselves, or other people, particularly well,” which makes learning interpersonal skills especially difficult.
That gap becomes more pronounced in professional settings. Markman points out that people are often taught what to do—deliver feedback, show empathy, lead a team—but not how internal states like motivation, fear, or cognitive bias shape behavior in real moments.
Emma Weber is the CEO of Lever, whose AI-based coaching tool focuses specifically on reflection and behavioral change. She sees this challenge reflected in what she calls the “knowing-doing gap.” Soft skills, she explained, are driven by internal factors such as “thoughts, feelings, values, beliefs, fears, and needs,” which makes them harder to transfer from training into sustained behavior.
In other words, practicing a skill once doesn’t guarantee it will show up when it matters. A manager might rehearse delivering constructive feedback in a training session, for instance, and still revert to criticism or avoidance the moment they’re under pressure with a real direct report.
This distinction matters when evaluating AI tools. Teaching someone what empathetic language sounds like is not the same as teaching them how to notice another person’s emotional state in real time.
Where AI can help: Feedback, contrast, and practice
Both experts agree that AI has real value when used in carefully defined ways, particularly around feedback and comparison.
Markman described a law school exercise that illustrates this well. Students wrote letters to clients, then fed those drafts into an AI system with a prompt asking it to rewrite the letter in a way that expresses more empathy and uses less technical language.
“Often what you saw in the rewritten letters was a little bit more of an expression of empathy for the situation that someone was in, as well as often a less technical vocabulary,” Markman said. “As long as you don’t offload responsibility for doing those things to the model, [the comparison can be] very powerful in teaching you about what information you’re not including in your original drafts.”
AI can also provide mechanical feedback that humans struggle to self-monitor. Markman pointed to speech patterns: upward inflection, according to media trainer Kim Dower, signals uncertainty in ways speakers rarely notice. Machine learning classifiers can detect and flag these patterns in real time, making them a natural fit for interview prep and leadership training.
Weber sees similar value in AI-facilitated practice, especially through role-play. Simulated conversations can help people rehearse difficult interactions and build confidence, but she stressed that practice alone is insufficient.
“Practice is not adoption,” she noted, adding that without reflection, people may perform well in training yet revert to old habits under pressure.
CEO of Lever
Reflection, not advice, as the engine of growth
Where Weber’s approach differs sharply from many AI coaching tools is in what her system doesn’t do.
Her AI, Coach M, is designed specifically to facilitate reflection. Rather than offering advice or solutions, it asks questions that push users to examine their own thinking and motivations.
Sessions are time-bound, typically 20–30 minutes, and users are guided to examine their own thinking rather than receive answers. The goal reflects both ethical and practical considerations, Weber explained, helping people to “have a better conversation with themselves.”
The system avoids prescriptive advice, reduces the risk of dependence, and preserves user autonomy. Weber added that tools must be “purpose-built” with deliberate constraints, especially when dealing with emotionally sensitive material: time limits, a no-advice policy, and a clear path to human support if needed.
Her data suggests this approach can scale responsibly. Out of approximately 24,000 sessions, only two required escalation to human intervention—a rate of roughly 0.008%. Both involved users who expressed signs of acute emotional distress. She points to this as evidence that constant human oversight is impractical and unnecessary for learning transfer.
The line between empathy and imitation
Still, both experts caution against mistaking imitation for understanding. Markman draws a clear distinction between learning empathetic language and developing empathetic attention. AI can help with the former, while the latter requires a shift in what people notice.
“One possibility [with AI coaching tools] is you learn to emulate empathetic speech,” he said. “That’s a start. It’s better than just bullying through your life.” But genuine empathy involves learning to “pay attention to other people’s situations” and “get outside your own head.”
This deeper form of learning, Markman explained, mirrors how expertise develops in any field: by becoming better at noticing relevant cues. Over time, reflecting on what the AI highlights—and asking why it didn’t occur naturally—can prompt that shift.
Weber echoes this concern from a different angle. She warns that mechanically applying AI-generated advice risks creating a kind of false connection where the language sounds empathetic, but the underlying understanding is missing.
AI as partner, not replacement
Despite their different perspectives, both experts describe AI as most effective when it supports human judgment rather than replacing it.
Markman described current models as “part of a balanced breakfast.” They can generate drafts, simulate interactions, and provide raw material for learning, but they lack a true model of human minds
Unlike a human coach, the AI has no genuine understanding of the person it’s talking to—it can only predict likely words, not perceive the emotional or motivational reality behind them. The most effective teacher, he emphasized, is still a human expert, particularly when learners have significant gaps in understanding.
Weber, meanwhile, emphasizes scale. Coaching has historically been expensive and inaccessible. The ICF’s global industry study puts the average one-hour coaching fee at $244 (and $272 in North America).
AI tools designed around reflection, she argues, can help broaden access by creating space for self-awareness and connection rather than focusing solely on mechanistic productivity.
So, can AI teach soft skills?
The answer depends on what we mean by teaching.
AI can’t feel empathy, understand motivation the way humans do, or replace the nuanced judgment of an experienced coach or mentor. But Weber’s 24,000-session dataset and Markman’s law school exercise both suggest it can meaningfully expand access to the kind of feedback and reflection that good coaching provides.
Used carefully, AI can help surface blind spots (by highlighting what people omit or overlook), create low-stakes environments for rehearsal, and prompt reflective questions that support learning. Used carelessly, it risks encouraging superficial imitation, emotional dependence, or forms of influence that users may not recognize or fully understand.
