I’ve been vocal about this on LinkedIn for months now, and I’ll say it again here because I think it’s the only starting point that matters: the future of our profession isn’t human coaching versus AI coaching. It’s an augmented model, human and AI together, and the sooner we stop treating this as a battle for territory, the sooner we can design something genuinely useful. There’s a lot of anxiety in the industry right now, about ethics, about quality, about whether coaches themselves are still needed. I understand where it comes from. But I don’t think fear is a good design principle. Human-centered design is. So let me walk through how I actually think about where AI belongs in a coaching relationship, and where, for now, it doesn’t.
Start with the boring stuff
When people hear “AI coaching,” they tend to jump straight to the dramatic version, an AI running a session. But that’s not where augmentation should start, and it’s not where it’s had the most value so far. The low-hanging fruit is bookings, matching, calendar coordination, re-bookings, all the unglamorous hygiene factors that quietly trip up a coaching journey before it’s even begun. Get those augmented and supported first. It sounds small, but a coaching relationship that starts with three re-schedules and a matching mismatch is already working uphill.
Where I draw the line
Once the hygiene layer is handled, the next question is whether AI should touch the actual content of the coaching conversation. I’m genuinely on the fence about recorded sessions, I think what happens in that room should stay confidential and private. But if note-taking is done well, that’s a legitimate place for augmentation: following up on action items, surfacing agreements, catching insights the client might otherwise lose by next week.
Where I’m not comfortable is the scenario where a human coach becomes the delivery vessel for an AI that’s actually running the session, proposing the questions, setting the direction, telling the coach what to say behind the scenes, scoring talk time and language. At that point the coach has been demoted from coach to voice. That’s the moment you strip the humanness out of coaching, and humanness is the thing coaching is supposed to protect.
Why I still think a human needs to be in the room
I wrote a post a while back pointing out that the ICF’s own definition of coaching doesn’t actually specify a human, which got me a fair amount of pushback from my own community. Fine, it’s worth probing. My honest answer is that a lot of what makes a coaching conversation work isn’t reducible to word choice. It’s the texture, the empathy, the felt experience of moving through something together. Even if AI eventually reads facial expressions and tone of voice, which would genuinely be remarkable, it’s still working from what’s said. A human sitting across from another human is working with more than that.
One example I keep coming back to: the silences a human coach can hold. Not a programmed pause of a set number of seconds, but the real-time decision, should I speak now or wait, that a coach makes because they’re reading a person, not a transcript. Counting seconds isn’t holding space. It just isn’t the same thing.
The value of a coach that’s awake at 2am
That said, I don’t think the case against AI running a session is a case against AI having a relationship with the client at all. One of the most genuinely valuable things about an always-on AI coach is exactly that, it’s on, at 2am, when someone has a real question and I am not picking up the phone unless it’s a distress call.
There’s also a disclosure effect that surprised me when I first saw the research. With a human coach, there’s still social currency at play, you want your coach’s esteem, you want to be seen well, even in a hard conversation. People will actually share the unvarnished, socially unacceptable version of a problem with an AI coach that they’d soften for a human. That gives an augmented relationship a second layer, an AI that captures the rawer signal, and a human who can synthesize the emotive, improvisational parts that AI still can’t reach. You don’t have to say please and thank you to it, and you don’t have to look good to it, and that turns out to matter.
Ethics without a supervisor to call
Coaches are trained to recognize the edge of therapy and step back. It’s a genuine question whether AI coaches know where that edge is. Most of the specialist AI coaches I’ve seen are built on ICF or British Psychological Society coaching standards, and machine learning is actually good at role definition, so those guardrails tend to hold. My bigger concern is generalist AI being used for coaching-shaped conversations with no role definition at all.
But here’s the honest part: this isn’t a new problem AI invented, it’s an old problem AI has amplified. Human coaches still struggle with boundary management. Coaching is therapeutic without being therapy, the boundary is real but permeable, and when a human coach hits that edge, they go to their supervisor and ask how to handle it. An AI coach can’t do that. It will have a hard on-off switch, this is outside my boundaries, or it won’t recognize the boundary at all. How well that’s handled depends entirely on who built it and what line of responsibility they took.
Coaching doesn’t travel the same way in every culture
Coaching AI is being built all over the world, and I think it inherits a cultural set point from wherever it was designed. Coaches trained in different markets carry different philosophies and different histories with them, and I’d expect AI coaches to carry the same bias, just less visibly.
Take power distance. In a high power-distance culture, a client might disclose less to a human coach because they want that coach’s approval, there’s a relationship being negotiated. An AI coach doesn’t trigger that dynamic in the same way, which could actually open up disclosure that a human relationship would hold back. Then there’s the deeper cultural blueprint: Europe and North America both carry a strong therapy tradition that shapes what people expect from a coaching conversation, more psychodynamic, more experiential, in one case; more GROW-model, action-oriented, cheerleading in the other. In parts of Asia, I’ve seen more of a mentor or advice-giving pull. An AI coach designed inside one of those traditions, meeting a user base shaped by another, might land beautifully or might just feel off. I don’t think we know yet. It’s something every organization bringing in an AI coach needs to ask, not just what language does this speak, but what cultural sense of coaching is baked into it, and how much can it actually adapt.
Cost, governance, and the real reason HR is buying this
Every buyer needs to start with the same question I started this piece with: what problem are we actually trying to solve, and is this the right tool for it, or just the newest one. Cost is a real driver, coaching AI can extend coverage to far more people, but cost and efficacy don’t always move in the same direction, and most organizations are still on a maturity journey figuring out whether wider coverage is actually translating into a better learning or development outcome.
The driver I think gets underestimated is governance. Human coaching has a black box problem, nobody really knows what happens in the room, one coach’s method looks nothing like another’s, and most companies aren’t measuring return on investment at even a basic level. AI coaching gives some organizations a sense, whether or not it’s fully proven yet, that they can leapfrog straight to something governable: one coaching model, one interface, consistent AI usage they can actually see and discuss. That’s a real opportunity. It also opens up a new set of challenges of its own.
What recruitment already taught us
I started my career in recruitment when it was still entirely analogue, CRM and cold calling and newspaper ads. I watched that industry move from high-street agency work to in-house functions to today’s agentic sourcing and screening tools. It automated fastest because the journey was clearly mapped and repeatable, and it hit the classic first problems of bias along the way, the well-known case of a hiring algorithm favoring ‘men in blue shirts who went to a particular sailing school’ being the famous example. The industry saw it, fixed it, and kept going, to the point that candidates now sometimes feel frustrated they never spoke to a real human in the process. It works for companies. It also became commoditized and, in places, dehumanized.
Coaching is a much younger industry moving through the same journey, fast. We solved the booking, screening, and matching problems first, the same low-hanging fruit I started with. Now we’re in the harder wave: how do we bring real value and automate the processes around a coaching journey without doing what recruitment did to itself. My guardrail is simple: keep the human coach in the loop. Automate everything around the two people in the room. Don’t automate the room.
Every one of these threads, the hygiene automation, the line I won’t cross on who runs the session, the always-on access, the cultural set point baked into design, the governance case HR is quietly building, comes back to the same design test I keep applying: does this serve the coachee, does it serve the coach, and does it hold up to why we started coaching people in the first place. Augmentation done well doesn’t replace the magic in the room. It protects the time for it to happen.
