Not every AI that can hold a conversation can hold a coaching conversation…

The distinction between what an LLM is capable of and what a properly constructed AI coaching platform is designed to do, is one that matters a great deal to us here at D4SP. We’re convinced that AI has a hugely beneficial role to play in contemporary coaching and we’ve therefore spent significant time and effort finding the right technology partner and investing in the sophisticated customisation necessary to produce a ‘proper’ AI coaching companion.

SOLAS, D4SP’s AI coaching assistant, is built on the Coachbot.ai platform; a framework developed specifically for professional coaching, not a general-purpose AI that’s been given a coaching-flavoured prompt and pointed in the direction of personal development. Coachbot.ai is structured around the ethical standards and competency frameworks of the EMCC, which means the principles governing how Solas engages with a user are grounded in the same professional standards that govern the practice of accredited human coaches. This is a critical difference between a system that’s been engineered to coach, and one that has been trained to provide all the answers.

The guardrails built into the platform matter for exactly the same reason; good coaching isn’t just about asking thoughtful questions, it’s also about knowing what coaching should not do. Solas knows (by virtue of its baked-in boundaries) when to refer, when to slow down, and when a conversation has moved beyond the appropriate scope of an AI coaching relationship. And its ethical framework isn’t a filter applied after the fact - it’s a fundamental part of the architecture. That distinction is important, particularly in situations where people may be navigating real difficulty, pressure, or genuine vulnerability.

When it comes to our users’ data, we’re equally unambiguous - conversations within Solas are confidential. They are not visible to employers or line managers, are never used to train underlying models, and are held to enterprise-grade standards of encryption and data residency. After all, a person cannot engage honestly in a coaching conversation if they’re uncertain about where that conversation data might end up. As with all coaching relationships, trust has to be a precondition rather than an afterthought, hence this level of data security is a foundational design principle of the coachbot platform.

None of this can be replicated simply by asking a general-purpose AI to “act like a coach.” You might get an interesting conversation (perhaps even a useful one) but you would not have the methodology, the ethical scaffolding, the professional guardrails, or the data protections that make AI coaching something an organisation can deploy responsibly, and an individual can engage with safely.

So, what does this mean in practice?

1. First, it means that the question organisations need to be asking themselves isn’t “can AI support our people?”, because the answer to that is almost certainly yes. The more important question is: “what kind of AI, built on what foundations, and governed by whose standards?”

2. Second, it shifts the question of accountability. When an organisation deploys AI coaching at scale, across hundreds or thousands of employees, including some who will bring real vulnerability into those conversations, the question of ethical architecture is critically important. If the platform hasn’t been designed with professional boundaries in mind, the gap between “useful tool” and “inadvertent harm” can close faster than anyone expects. HR leaders and L&D professionals who are serious about duties of care need to understand what’s under the bonnet, not just what’s on the bot landing page.

3. Third (and this is the point that I find most interesting) a properly designed and implemented AI coachbot reframes what AI coaching is actually for. The right kind of AI doesn’t replace human coaches; it extends the reach of coaching to people who would never otherwise be able to access it; the manager navigating a difficult team dynamic at 10pm or the employee who won’t ask for help in a group setting but might do so (quietly) with the right kind of structured support. AI coaching, done properly, democratises something that has historically been a privilege only for senior managers.

As AI coaching becomes less of a curiosity and more of a mainstream tool, the organisations that get this right won't just have deployed a technology; they'll have made a deliberate choice about the kind of support they're prepared to offer their people, and the standards they're willing to hold themselves accountable to. In my humble opinion, that choice deserves far more scrutiny, backed by far more ambition, than most AI tool procurement decisions currently receive.