I recently had one of those rare conversations that makes you feel more certain about the choices you’ve already made. She’s an AI researcher working at the intersection of machine learning and societal impact. She was with Google and DeepMind (one of the 100 original working on Gemini) and is now in academia in Europe studying AI evaluation, fairness, and the risks of AI companions. Her father, a manager to a participant in one of our cohorts, connected us. We ended up talking for an hour, and what emerged was a shared language and clarity about AI coaching limitations and how or if AI will intersect with QuantumBloom’s offerings.

Here’s what we got into.

Why AI Coaches Can’t Be Held Accountable

When people ask whether AI will replace coaches, they’re usually asking an economic question. Is it cheaper? Is it faster? Can it scale?

Those aren’t the wrong questions. But they’re not the right ones either.

The questions QuantumBloom keeps coming back to are: Is it better? Who is accountable?

Our coaches are certified. They’ve completed hundreds of hours of supervised training. They operate under a code of professional standards. When one of our early-career women in STEM is trying to decide whether to surface a harassment incident, maybe factoring in a toxic manager, a fragile team dynamic, and her own financial risk tolerance, her coach knows her. Her coach has context. And critically, her coach is accountable to her as a human being, not to a product roadmap.

Chatbots cannot take accountability for themselves. A company that built an AI coach has a very different relationship to its users than a human coach does and thus far, seems to be distancing itself from any form of implied accountability. There’s no reciprocity. There’s no skin in the game. And if the advice goes wrong, if someone follows “best practice” without the context that makes best practice meaningful, who answers for that?

Right now, the answer is no one.

AI Coaches are Designed to Validate, Not Challenge You

Off-the-shelf language models are tuned to be overly validating. It’s a design choice to make people feel good about their chatbot relationships, which in turn drives the engagement companies want and need.

That is the opposite of what good coaching does. Real coaching is about helping someone understand who they are, what they value, and how to operate authentically in a system that may be actively working against them.

Sometimes that requires building someone up. Sometimes it requires tough love. Sometimes it requires sitting with someone in a moment of uncertainty and not rushing to a resolution. An LLM calibrated for user satisfaction cannot do that because doing it well means occasionally making the user uncomfortable.

LLMs Engineer Away the Wrong Kind of Friction

There’s a broader cultural dynamic worth naming here. Technology has spent two decades removing friction from every interaction. Faster checkouts, smoother onboarding, one-click everything. And often that’s fine.

But we’ve started applying the same logic to human development, and that’s where it breaks down. Difficulty, discomfort, and failure build professional skills. They are the process. You don’t learn to have a hard conversation with your boss by reading about it. You learn by having it, getting it wrong sometimes, recovering, and trying again.

An AI coach optimized for ease fails to replicate that and, worse, it actively works against it. If every interaction is frictionless and validating, you’re not building resilience.

The messy, uncomfortable, deeply human work of figuring out who you are in a workplace that wasn’t built for you cannot be friction-engineered away. It has to be lived.

What AI Coaching Misses About Human Connection

The researcher described QuantumBloom’s model as instilling social fabric. Not a term I use, but one that certainly describes our multi-prong, multi-touchpoint model.

Yes, QuantumBloom is building social fabric. An 18-month experience where participants form real communities with each other and their coaches, managers, and mentors. They surface their learning in their actual work contexts. They watch each other grow. They become, genuinely, each other’s support systems.

A chatbot cannot be part of that because chatbots aren’t part of any social fabric. They don’t exist in the world. They have no experience of their own. They encode patterns from millions of people’s accounts of their experiences and what they reliably produce is the most generic version of all of it.

Why AI Coaches Provide Generic Advice to Non-Generic People

LLMs are trained to recognize patterns in data, and they reliably trend toward whatever is most common, most generic, most average across their training set.

Deploying AI coaches would be a profound mismatch with the work we do. Our participants aren’t having average experiences. They are naive employees navigating a workplace in flux. Their experiences are, by definition, outside the norm the model was trained on.

So the very populations who most need nuanced, contextually-aware support are the ones least well-served by a system optimized for the generic case. An AI coach would give them the average of everyone else’s experience, handed back as advice, which is more noise than support.

Why Coaching Outcomes Can’t Be Optimized Like Software

AI systems excel when they can check their own work. Code either runs or it doesn’t. Data either matches or it doesn’t. But coaching outcomes like Am I more confident? Do I feel like I belong here? Am I more likely to stay in this career? resist that kind of clean feedback loop.

We’re doing everything we can to bring rigor to our measurement including, baseline surveys, interim surveys, matched-pair analysis, validated scales across nine outcome categories, statistical testing. We have researchers involved. We take this seriously.

But even we can’t always isolate which piece of the experience drove which outcome. Was it the group coaching? The one-on-one sessions? The manager engagement? The peer community? The curriculum? Probably all of it, interacting in ways that are hard to untangle.

The layered, multi-modal, deeply human nature of what we do is exactly what creates the effect. An AI-optimized version of it (clean, measurable, scalable) would strip out the messy parts that are actually doing the work.

Before Buying an AI Coaching App, Ask This

As QuantumBloom continues to keep an eye on AI, we will keep asking What problem would an AI coach actually solve? Not what problem could it theoretically solve.

For most of the spaces AI is being proposed as a solution, that question doesn’t have a clean answer yet. The technology is moving fast, the hype is moving faster, and the companies behind these models need adoption to justify their economics. No guardrails. Just a race to profits, a race to first.

For QuantumBloom, the gap we’re closing is human connection in increasingly AI-driven environments. That’s a people, not tech, problem and we intend to keep solving it with people.