Here is a conversation happening right now in boardrooms, coaching communities, and HR departments that nobody is having cleanly.
In boardrooms, the question sounds like this: "If an AI can coach our managers at a fraction of the cost, why wouldn't we?" In coaching communities, it sounds like this: "I'm not worried about AI" — said in the tone of someone who is very worried about AI. Both conversations are circling the same elephant. Nobody wants to name it precisely because naming it requires having an actual answer to it.
I want to offer that answer. Not as reassurance, and not as hype — but as science.
In May, I had the privilege of presenting at ICF International Coaching Week alongside my colleague Gabriela Savu, ACC, to an audience of 41 coaches from six countries. What I observed in that session confirmed something I have been thinking about for over a decade: the reason AI cannot replace human coaching is not philosophical, it is empirical. And the science that explains it is not coaching science. It is motivation science — the kind that matters equally to coaches and to the leaders who decide whether to invest in them.
The Research Is Real. So Is the Gap.
Let me start with what the evidence actually shows, because both sides of this debate get it wrong.
AI coaching works — for specific things. A rigorous randomized controlled trial over ten months (Terblanche et al., 2022) found that an AI coaching chatbot matched human coaches on structured goal attainment. If you are a CEO evaluating AI coaching platforms, that finding is real and you should not dismiss it. AI excels when the task is structured, process-based, and goal-oriented: accountability check-ins, progress tracking, between-session reflection prompts, habit formation. There are platforms built for exactly this that deliver measurable value.
But here is what the same research reveals, and what most sales decks quietly skip: when clients are asked to evaluate human coaches versus AI coaches after a single session — measured on insights, working alliance, commitment, trust, confidentiality, stress reduction, and sense of being truly heard — human coaches score higher on every single dimension. Not most. Every one. (Passmore et al., 2025)
The pattern is too consistent to be coincidental. It points to something structural, not circumstantial. And it took me fifteen years of studying motivation science to understand why.
The Science Nobody Teaches Coaches — and Every Leader Should Know
Self-Determination Theory is a 40-year, cross-cultural, empirically replicated research program developed by professor Deci and professor Ryan that has become one of the cited frameworks in psychology. Some coaches have never heard of it by name — even though they practice it every session. Many leaders have also rarely heard of it — even though it explains most of what they observe about why their people are or are not engaged.
The central finding is this. All human beings are born with three basic psychological needs. These are not preferences, not personality traits, not cultural values. They are needs in the same sense that food and sleep are needs: when they are satisfied, people experience vitality, growth, and autonomous motivation — the deepest, most durable form of engagement. When they are frustrated, performance declines, well-being erodes, and behavioral change does not hold.
The need for autonomy is the felt sense that my choices come from me — that my actions are expressions of my values rather than responses to external pressure. This is not independence. It is volition within structure. Children show this need from their first months of life, and executives show it in every decision they resist because it was not theirs.
The need for competence is the experience of growing effectiveness. Not external validation, not ranking — the internal felt sense of I am capable of this. When a manager finally solves a problem they have been circling for weeks, the energy they feel is competence satisfaction. It is as physiologically real as anything on a blood panel.
The need for relatedness is the experience of genuine connection — being cared for, being seen, belonging to something beyond oneself. This is the most misunderstood of the three. It is not warmth, rapport, or likability. It is the deep sense that the other person is actually with you: present, caring, without agenda.
When environments satisfy all three needs consistently, people become intrinsically motivated. They bring discretionary energy. They take risks, adapt, and grow. When organizations frustrate these needs — through micromanagement, performance culture built on fear, relational transactionalism — they produce exactly what most leaders complain about: disengagement, compliance-without-commitment, and talent that walks as soon as the market improves.
This is not idealism. This is documented organizational science.
Why This Is the Answer to the AI Question
Here is the insight that changed my thinking — and the one that shifted something for the coaches in that room.
AI can partially support autonomy. Personalization, choice architecture, non-judgmental 24/7 access — a well-designed coaching app can create genuine conditions for autonomous choice. AI can also support competence: structured feedback loops, progress tracking, data-driven insight. These are real contributions to the coaching process.
But AI cannot create relatedness. Not today. Possibly not ever.
Relatedness requires intersubjectivity — one consciousness genuinely meeting another. It requires the client to know, below the level of rational processing, that the entity across from them is a person who actually cares.
In the early 2010s, the University of Southern California — with DARPA funding — developed a virtual interviewer called "Ellie" to help screen returning soldiers for PTSD symptoms. The finding surprised many: soldiers disclosed significantly more about their mental health to the avatar than they did on official military assessments. The reason they gave was consistent: they felt less judged by something that wasn't human. That is a real and important finding for AI in any helping profession. But Ellie was a screening tool, not a therapist. And the researchers were careful to say so. The moment you move from identifying a problem to working through it — from recognition to transformation — the relational requirements change entirely.
The data from Passmore et al. now shows the same pattern in coaching. Stress reduction, trust, the sense of being truly heard — these do not emerge from process competence. They emerge from the experience of genuine human connection. Which means they emerge from relatedness. Which means they require a human.
This is not coaching's soft, unquantifiable value. This is its only irreplaceable one. And it maps directly to the most powerful need in the SDT framework.
A Practical Framework for Coaches and Leaders
Understanding this does not require abandoning AI — it requires deploying it correctly. After working with coaches and leaders on this, I use a simple three-zone model.
Delegate to AI: Scheduling, progress tracking, session note summaries, between-session reflection prompts, data aggregation, assessment scoring. These take your time without using your humanity. Give them away.
Collaborate with AI: Session preparation, identifying client patterns over time, generating reflection questions, scaling your reach. AI as a thinking partner — with you, not instead of you.
Protect as yours alone: Trust-building, holding emotional space, navigating ambiguity, co-creating meaning, authentic presence, ethical judgment, the relational bond. This is not a list of things AI cannot yet do. It is a list of things AI will not do — because they require consciousness meeting consciousness.
For leaders evaluating AI coaching platforms: the platforms that show the best outcomes use AI to extend human coaches, not replace them. The ones that remove the human entirely are buying the structured goal attainment and leaving the transformation on the table.
For coaches: the question is not whether to integrate AI into your practice. It is whether you know precisely which parts of what you do are yours alone — and whether you are investing in deepening those parts or hoping the rest goes unnoticed.
The Conclusion No One Is Drawing
I want to end with something I said in that session that I believe deserves to be said more widely.
The panic statements coming from certain AI executives — jobs will disappear, professionals will be replaced — deserve scrutiny. Fear is excellent fundraising. Whether it is excellent analysis is a different question.
What the science shows is more nuanced and, I would argue, more interesting: the capabilities that generate lasting transformation in human beings — genuine presence, authentic care, relational depth — are not lagging behind AI development. They are categorically different from what AI does. A calculator did not replace mathematicians. It changed what mathematicians do. Something similar is happening here, and the coaches and organizations that understand it early will be well positioned.
There was a moment near the end of our ICF ICW session when a participant, a coach with a background in technology, said something that has stayed with me:
"There is enough room for us. The essential part is mastering humans. This touch of humans."
She was right. And the science agrees.
The algorithm can optimize. Only you can connect.
That is not a motivational statement. That is a finding.
This is the third edition of The Pink Elephant Papers — where we put on the table what leaders see but don't say. I am grateful to Romania Chapter of ICF for the platform and to Gabriela SAVU, ACC, for her partnership in delivering the original session this article draws from.