Artificial intelligence is being called everything from humanity’s salvation to its downfall. The reality is more nuanced, and more human. The people already thriving with AI share something more important than access: they share an approach.
In the past two years, a familiar pattern has emerged in workplaces, universities, homes, and creative studios around the world. Two people sit down with broadly similar tools (a large language model, an image generator, a code assistant) and have completely different experiences. One finds it transformative. The other finds it underwhelming, frustrating, or worse, actively damaging to the quality of their work. The difference is rarely about intelligence or technical expertise. It is about approach.
The conversation about AI and the future of work has been dominated by questions of access; who has it, who doesn’t, what it costs, and whether it will replace human labour. These are genuine and important questions. But they risk obscuring a subtler and perhaps more consequential issue: among those who have access to AI tools, the gap in outcomes between those who use them well and those who don’t is widening rapidly. And that gap is shaped by education, confidence, and what researchers are beginning to call ‘AI literacy’.
The Prompt Is the Skill
The interaction between a human and a large language model is mediated almost entirely by language, specifically, by the quality of the instruction, or ‘prompt’, that the human provides. A vague or poorly constructed prompt yields a generic, often mediocre output. A precise, contextualised, and iterative prompt can yield something remarkable. This seems simple. In practice, it is a skill, and like most skills, it is unevenly distributed.
Research conducted with workers across multiple sectors has found that those who benefit most from AI tools tend to share certain characteristics. They treat the AI as a collaborator rather than an oracle. They provide context, ask follow-up questions, push back on outputs they find inadequate, and iterate. They have a clear sense of what they are trying to achieve and are able to evaluate the AI’s output critically rather than accepting it uncritically. This final skill, critical evaluation, turns out to be perhaps the most important of all.
Those who benefit least, by contrast, often approach AI with a kind of anxious passivity, either expecting it to do everything for them, or so concerned about being ‘wrong’ that they accept whatever it produces. Research on students using AI tools for writing and research has found that those who use AI outputs without significant critical engagement tend to produce work that is worse than what they could produce alone, flatter, less distinctive, less accurate.
The AI Confidence Gap
One of the most consistent findings in research on AI use is the existence of what might be called an AI confidence gap. Women, on average, report lower confidence in using AI tools than men, even when their actual competence is comparable or superior. People from lower socioeconomic backgrounds report higher anxiety about using AI ‘incorrectly’. Older workers report feeling behind a technological curve they did not choose to run on.
These confidence gaps are not trivial. They shape how people approach AI tools, how much they experiment, and how willing they are to advocate for themselves when AI systems produce problematic outputs. They are also self-reinforcing: lower confidence leads to more passive use, which leads to worse outcomes, which confirms the belief that AI is ‘not for them’.
Those who benefit least from AI often approach it with anxious passivity, either expecting it to do everything, or so concerned about being wrong that they accept whatever it produces.
How Different Communities Are Actually Using AI
The deployment of AI across different communities and sectors reveals a rich and often counterintuitive landscape. In healthcare, doctors and nurses are using AI tools for everything from literature review to administrative tasks, but the quality of those tools, and the trust placed in them, varies dramatically by institution and by specialty. In creative industries, AI is being used both to accelerate and augment human creativity, and to replace it, and the same tool can do either, depending on how it is deployed.
In education, the picture is particularly complex. Teachers in well-resourced schools are using AI to personalise learning, generate differentiated materials, and reduce administrative burden. In under-resourced schools, AI is arriving with less professional development support, higher anxiety, and, in some cases, as a shortcut rather than an amplifier. Students who are already strong writers and thinkers tend to use AI to become stronger; students who are struggling may use it in ways that deepen rather than address their difficulties.
Small business owners represent one of the most genuinely exciting AI success stories. Entrepreneurs (particularly those from communities that have historically lacked access to professional services like legal advice, marketing expertise, or financial analysis) are using AI to access a kind of affordable, always-available expertise that was previously out of reach. The caveat is that the quality of what they get depends heavily on what they ask for, and on their ability to verify what they receive.
The Question of Who Designs These Tools
The approaches that work best with current AI tools reflect, to some degree, the assumptions of their designers; overwhelmingly people who are confident, highly educated, and comfortable with iterative, experimental processes. This is not a neutral design choice. It encodes particular cognitive styles as the ‘right’ way to interact with AI, and it disadvantages those whose communication styles, learning histories, or cultural backgrounds lead them to different kinds of interactions.
As AI tools become embedded in more consequential areas of life (healthcare, education, employment, legal services) the question of how they are designed, who can use them most effectively, and what support is available to those who cannot, becomes a question of justice. Access to the tool is the beginning of the conversation. The quality of what that access actually delivers is where the real work starts.