AI Confidence Is Soaring. AI Competence Is Not Keeping Up.

The most dangerous person in your office right now is not the one who refuses to use AI. It is the one who uses it constantly and has no idea when it is lying to them.

We are living through a strange and underreported moment. Adoption of AI tools has exploded across virtually every profession. Teachers are using it to build lesson plans. Lawyers are using it to draft briefs. Marketers are using it to generate campaigns. HR managers are using it to write job descriptions and screen candidates. And in almost every case, confidence in these tools is running several miles ahead of actual competence with them. That gap, quiet and widening, is where the real risk lives.

The Tool That Sounds Certain Even When It Is Wrong

Here is a scenario that is no longer hypothetical. A paralegal uses an AI assistant to research case precedents for a motion. The output is fluent, authoritative, formatted perfectly. It reads exactly like something a senior attorney would produce. She submits it. The attorney files it. The judge notices that two of the cited cases do not exist. They were fabricated by the AI, a phenomenon researchers call hallucination, with complete stylistic confidence. This is not a cautionary tale from the early days of ChatGPT. Versions of this story have played out in actual courtrooms, with actual consequences for actual professionals.

The problem is not that AI makes things up. We can learn to work around that. The problem is that AI makes things up in a voice that sounds nothing like uncertainty. There is no hesitation, no hedge, no tell. When we are wrong about something, we usually sound a little less sure. AI does not have that signal. It presents a hallucinated legal case with the same smooth confidence it uses to summarise a real one. And because most of us are using these tools precisely because they save us time, we are not always stopping to check.

This is what we mean by the competence gap. Confidence in AI is partly driven by the quality of the interface, the fluency of the output, the sheer relief of having a capable-seeming tool at our fingertips. Competence, the ability to interrogate that output, to know when to trust it and when to verify it, is a different skill entirely. And it is one that almost no organisation is formally teaching.

Why This Is Everyone's Problem, Not Just the Tech Team's

It would be easy to frame this as a problem for specialists. Train the IT department, build some guardrails, let the engineers sort it out. But that is exactly the wrong instinct. The people most exposed to this gap are not the technical staff. They are the professionals in every field who are now using AI as a daily work tool, often without training, often without organisational guidance, and often without a clear sense of what the tool's failure modes actually look like.

A nurse using an AI tool to help interpret symptoms and flag potential drug interactions needs to know not just how to use the tool, but how to recognise when the tool is confidently wrong. A small business owner using AI to draft contracts needs to understand that fluent language is not the same as accurate language. A teacher using AI to generate quiz questions needs to know that the tool can produce plausible-sounding historical facts that are simply false.

Competence here is not technical. We are not talking about understanding how large language models work at a mechanical level. We are talking about a set of professional instincts: healthy skepticism, verification habits, an awareness of the specific ways these tools tend to fail in your field. That is learnable. But we have to decide it is worth learning.

The organisations that are getting AI adoption right are doing one thing differently from everyone else. They are not just asking whether their people are using AI. They are asking whether their people understand what AI cannot do. They are building cultures where it is not embarrassing to say, I ran this through the AI but I want a second pair of human eyes on it. They are treating AI competence as a professional responsibility, not a personality trait of the curious.

The question we each need to sit with is this: in our own work, are we confident because we understand the tool, or confident because the output looks good? There is a meaningful difference. And right now, in most of our workplaces, nobody is asking us to tell them apart.