What would AI coaching actually have to do to deserve the name?

The internet is full of tools that dress up keyword-matching and pre-written responses in the language of personal development, and if that were all AI coaching amounted to, I’d be very disappointed. However, the reality is that well-engineered AI systems are capable of going way beyond such tools and can deliver truly transformational coaching experiences!

So, I want to be specific about what I mean by ‘AI coaching’, and equally specific about what I don't. AI coaching, done well, is not a scripted conversation tree, a mood-tracking app, or a digital version of the motivational content that already fills everyone's feed. It is a responsive, personalised experience that draws upon the same foundational principles as human coaching - active listening, powerful questioning, reflection, and accountability - and delivers them in a way that is available when a person actually needs them.

What it is not, and this point matters enormously, is a replacement for human coaching. The skilled human coach brings things that remain genuinely irreplaceable; deep relational engagement, professional judgement (developed over years of practice), and the kind of presence that changes the energy of a conversation. Many coaches (me included) are now clear that the optimal benefits from coaching are to be derived from a well-integrated amalgamation of expert human coach and well-designed AI coaching assistant - it’s really not an ‘either/or’ debate. But of course, for people lower down the organisational hierarchy, the choice has never been between AI coaching and human coaching; for most of them, there have simply been no coaching options at all!

That’s the gap that D4SP is building AI for; not so much for the executive with a quarterly coaching session already in the diary, but for the team leader promoted six months ago who is quietly struggling and has nowhere to take that struggle. And for the early-career professional making consequential decisions without a framework for thinking them through, or the manager who knows something in their team is off but has never been given the tools to address it. For those people, the question of whether AI coaching is as good as the best human coaching is, respectfully, the wrong question. The right question is whether it is better than what they currently have...?

So, what does AI “coaching” actually need to provide to merit that descriptor?

As a minimum, it needs to do what good coaching has always done: create the conditions for a person to think more clearly, act more deliberately, and develop a more accurate understanding of themselves and the impact they have on others. It needs to ask better questions than the person would ask themselves, hold them accountable to the commitments they make, and adapt to where they are rather than where a pre-written programme assumes them to be. It needs to be available at the moment of need, and it needs to know the difference between what a person is asking and what they actually need to explore.

These are not modest requirements - they require much more than a standard LLM - but they are, however, the right ones and they are the standard against which D4SP believes any serious AI coaching offer should be measured, including our own.