AI or IRL: CoachBots vs Professional Coaching

 Short on time? Listen to the AI-powered audio version—no scrolling needed.

What My First AI “Coach” Taught Me—And Why Real Coaching Still Needs Humans

In 2022, I had the opportunity to work on the inaugural AI products for Shopify. Prompt engineering was just becoming a term, and those of us tinkering with it were still in the early stages of figuring out how to get large language models (LLMs) to produce consistent, helpful, and reliable outputs—especially in a product experience meant to support merchants managing their backend systems.

Back then, we were working with early versions of ChatGPT-3.0, where the outputs were... well, unpredictable. Hallucinations—where the AI would confidently fabricate facts or mash together unrelated ideas—were common. Human logic had to do a lot of heavy lifting to course-correct the nonsense.

Another challenge, then and now, was context. LLMs don’t truly understand language; they predict it. They generate text by stringing together the most statistically probable sequence of words. So while “airplane” and “sky” are likely to appear together, “airplane” and “tuba” are not. It’s how you end up with outputs that sound natural—but only when the context aligns with the data the model was trained on.

Still, you'd think something as basic as synonyms would be simple enough. Not quite. One test I vividly remember from those early Shopify days involved a fictional women’s clothing brand. Every time we prompted ChatGPT to write product copy for “nightgowns,” it came back with descriptions fit for Cinderella headed to the ball—not something more down-to-earth like what Old Mother Hubbard might wear to bed.

So—what does all of this have to do with coaching?

Plenty.

Because while I may have been experimenting with a cutting-edge LLM back in 2022, it wasn’t my first encounter with the quirks of natural language processing.

Let’s rewind to the 1980s. I grew up within biking distance of my local science and technology museum, which had a small exhibit of computer terminals that fascinated me. One summer, I found myself completely captivated by a particular program—so much so that I spent the better part of my vacation pedaling to the museum to interact with it.

That program? ELIZA.

Created in the mid-60s by MIT’s Joseph Weizenbaum, ELIZA was one of the earliest attempts at simulating conversation through simple pattern matching. It was designed to mimic a Rogerian therapist, reflecting statements back to users with open-ended prompts like, “Tell me more about that.” It was also, in its own way, an early prototype of how non-directive Ai coaching frameworks operate.

Of course, I wasn’t using ELIZA for self-help. I was just fascinated by how it mimicked natural conversation. Every visit was an experiment: What could I say to keep the dialogue coherent—and how long until the cracks showed? “Tell me more about the giraffe in your family,” was usually the kind of moment where illusion gave way to code.

If you’re curious, you can still try a version of ELIZA online.

Looking back, those early interactions with ELIZA feel strangely prophetic. They helped shape my curiosity about language, tech, and what it means to “connect.” But they also laid the foundation for why I don’t believe AI will ever fully replace coaching.

Why? Because even Weizenbaum himself named the program after Eliza Doolittle—the character from My Fair Lady who learns to say all the right words but still isn’t accepted as truly transformed. Just like Eliza Doolittle, AI can mimic the surface of a human exchange—but it doesn’t understand the nuance, timing, or non-verbal signals that give real conversations their weight and meaning.

And that matters.

In fact, the stakes get higher when you consider the rise of CoachBots and TherapyBots. While coaching isn’t therapy—and isn’t meant to replace it—TherapyBots offer a relevant comparison point, as many coaching bots draw from similar conversational frameworks. A 2025 study from Stanford, “Expressing Stigma and Inappropriate Responses Prevents LLMs from Safely Replacing Mental Health Providers,” April 25, 2025, (paper, GitHub) found that even when faced with clear indicators of emotional distress, TherapyBots routinely missed the mark. In one case, a bot responded to a depressed user by continuing to list the heights of Manhattan’s bridges—completely bypassing the emotional context of the conversation.

Worse, most consumer-grade TherapyBots performed poorly overall, with fewer than 50% of questions answered appropriately. Caveat emptor indeed. 

Context is everything

Half the strategy of radical leadership is understanding context, navigating nuance, and tapping into shared human experience. And that’s exactly where LLMs fall short.

In coaching—especially executive or leadership coaching—clients operate within high-stakes environments. They manage large budgets, lead teams, and are responsible for decisions that affect hundreds, sometimes thousands of people. That requires more than AI-generated reflection or pattern recognition. It requires ethical grounding, emotional presence, contextual fluency, and a skilled integration of diverse coaching modalities that go beyond non-directive exploration—tools that adapt to complexity, support transformation, and meet the moment.

That said, I don’t believe it’s CoachBot versus coach. It’s about what’s fit for purpose.

CoachBots can be incredibly effective for self-driven goal-setting. When someone already knows what they want and needs help breaking it down into steps or staying motivated, a well-designed AI companion can help. In larger organizational settings, CoachBots can also provide scalable insights—surfacing patterns, identifying common issues, or offering quick check-ins that human coaches alone couldn’t possibly manage at scale.

And perhaps most meaningfully, CoachBots offer accessibility. They have the potential to democratize support across socioeconomic lines—something that’s long overdue in the professional development space.

A CoachBot prompts. A human coach partners.

When AI first landed at Shopify, many of the brightest minds were grappling with the same fears I hear today: Will AI replace us?

But as one of my colleagues wisely pointed out, the ATM didn’t replace bank tellers—it just changed the nature of their work.

That’s why I continue to see AI as an “and”—not an “or.” I believe the future of coaching lies in integration: CoachBots and human coaches working in tandem as part of a more holistic, sustainable approach to leadership development.

The International Coaching Federation defines coaching as a partnership—a creative, thought-provoking process that supports people in becoming their best selves. And while AI might simulate aspects of that partnership, the real breakthroughs—the “a-ha” moments that move people forward—are born from connection. From trust. From presence.

True partnering means two (or more) people working together toward a shared goal. It means seeing and hearing each other fully, something a CoachBot simply isn’t designed to do. At best, it can mimic. At worst, it misses the moment entirely.

Just like ELIZA did, all those years ago.

This is the heart of radical leadership: staying awake to complexity, holding space for nuance, and refusing to outsource what makes us most human—especially in moments that matter. As technology becomes more integrated into how we work, lead, and grow, we’re faced with an essential choice: to lean into convenience, or to double down on connection. Because if leadership is fundamentally relational, how do we keep the human at the center of our growth?

To your success!

Curious how a human coach can partner with you to support your growth in ways Ai can't? Book a free coffee 1:1 and let's strategize together on what that might look like for you. 

Written and edited by: Kristina Mausser
Review generated by: ChatGPT, OpenAi. Revised for clarity.