Core idea: Companies are rolling out powerful AI tools and hoping employees will figure them out alone. But access is not adoption. AI only becomes useful once people are coached past the discomfort of talking to a machine — and learn to turn everyday information into something the organisation can reuse.

Agile is a simple idea. Work in short cycles, prioritise ruthlessly, gather feedback, improve continuously — the whole method fits in a paragraph. Yet companies still hire Agile coaches, Scrum Masters and transformation teams to make it stick. That is because understanding an idea and changing behaviour are two different things. Firms that learned this with Agile rarely apply the same lesson to AI: they hand out licences and expect adoption to follow on its own.

Businesses are licensing Copilot, ChatGPT, Claude and similar tools at remarkable speed, then leaving employees to explore them unaided. The assumption seems to be that an intuitive interface guarantees intuitive use. It rarely does. Most employees still use AI for little more than drafting emails or summarising documents, when the same tools could analyse projects, test assumptions, prepare reports and preserve institutional knowledge. The gap is not in the deployment. It is in what happens — or fails to happen — afterwards.

The technology itself is not hard to use. Anyone can write, speak or upload a document, much as they would in ordinary life. The harder problem is psychological. People often do not know what to ask, how much detail to give, or whether to trust the answer. Talking to a machine feels strange and slightly exposing — more so at work, where mistakes carry consequences and a person, not the machine, remains responsible for the outcome.

Call this the awkwardness barrier. It is not dramatic fear; it is a small hesitation that keeps people inside safe, trivial uses. They ask less than they could, share less than the task needs, and treat AI as a novelty rather than a working tool. The safest habit is also the least useful one.

Research on technology adoption backs this up. Studies on "technology self-efficacy" link confidence and organisational support to how readily people take up new tools. Early work on coaching around Microsoft Copilot points the same way, and workforce-training data shows that even a genuinely useful tool can sit idle without proper onboarding. The lesson is practical: a useful tool does not create useful habits by itself.

Coaching should start by changing the mental model. AI is often marketed as a synthetic personality — a digital colleague or assistant. That framing makes the technology sound friendly, but it can deepen the awkwardness: some employees feel asked to form a relationship with a machine, others fear it is here to replace them. A more useful model is to treat AI simply as an intelligent way of processing information.

The point is not that the system is human, but that people can talk to it in ordinary, human ways — speaking, writing, uploading a document, giving context — and it turns that material into something that can be organised, checked and reused. Ordinary communication has simply become a way of making work processable.

Take a manager running a weekly team meeting. Without coaching, she might use AI now and then to draft an email or summarise a document, but the meeting itself still runs on memory, handwritten notes and personal interpretation. Decisions get recorded selectively. Different people remember action items differently. Weeks later, the context has faded and the reasoning behind a decision is hard to recover — the work moves on, but much of what was learned along the way is quietly lost.

With coaching, the same manager uses AI to capture the meeting properly: decisions logged, actions assigned, discussions made searchable. AI helps track commitments, flag unresolved issues and produce a consistent record, so direction stays clear and follow-up is precise. Information that would once have been lost to memory and scattered notes instead becomes a reusable organisational asset, available the moment it is needed.

The technology has not changed. The behaviour has. In the first case, information is temporary; in the second, it becomes an asset — decisions, assumptions, risks and commitments that can be retrieved and reused long after the meeting itself is forgotten.

This is why AI's memory matters strategically. Human memory is selective: details fade, conversations blur, context disappears. AI can hold and retrieve far more authorised information than any one person could reliably remember. That does not make its judgement better, or every output correct — it makes AI valuable as a repository and processor of business information, provided its use is properly governed.

Coaching should teach one simple principle: capture once, structure consistently, reuse intelligently. A meeting should not vanish when the call ends. A voice note should not stay a forgotten recording. A chain of emails should not become buried context. With the right permissions, all of this can become decisions, tasks, reports and inputs for future automation.

This is the strongest case for training. Employees are not just learning to ask AI for text — they are learning to turn work into durable information: externalising thought, preserving context, creating material that compounds over time. Adoption becomes less about clever prompts and more about building organisational memory.

A caveat matters here. Not every conversation should be recorded, not every document uploaded, not every memory retained. Coaching must draw clear lines: what information is authorised, which sources can be trusted, what needs human checking, and who stays accountable for the decision. The goal is confident adoption, not uncritical trust.

Companies should also be wary of measuring AI success by licences issued or messages sent. Those are activity metrics, not adoption metrics. A better test is whether employees use AI to cut repetitive work, preserve useful context, sharpen decisions and tackle tasks that were previously out of reach. If a company deploys the technology, it also takes on responsibility for helping people understand where it fits into their jobs.

Most debate about artificial intelligence focuses on what the models can do. But organisational change rarely comes down to capability alone. Agile needed coaches not because its ideas were complicated, but because new habits take practice and reinforcement. AI will be no different.

The interface is already simple: talk, write, or hand over a document. The harder task is helping people feel comfortable enough to start, disciplined enough to keep going, and thoughtful enough to know what should be kept, processed and trusted. AI deployment supplies the tool. Coaching is what makes it useful.