Every organization is investing in AI literacy. Almost all of them are teaching the wrong thing.

Every L&D budget in 2026 has a line item for AI literacy. Every university has introduced some version of an AI skills module. Every corporate training catalogue has a course on prompting, on responsible AI use, and on integrating tools into workflow. This is not a bad investment. It is an incomplete one.

The thing being taught is mostly operational: how to write prompts, how to check outputs, how to use specific tools for specific tasks. The thing not being taught is what determines whether any of that produces anything worth having. The prompt is not the skill. The thinking behind the prompt is the skill.

The Gym Membership Problem

Imagine paying a significant amount every month for a gym membership, then hiring a personal trainer to attend sessions on your behalf while you stay home. You have invested in the infrastructure of fitness. You have not done any of the work that produces it. The muscle does not appear.

The membership is already paid for. Microsoft’s Work Trend Index 2024 found 85% of Korean AI users bring their own tools to work without company support, while only 39% of AI users worldwide have had any training from their employer. People are using the tool. Almost nobody is being taught to think with it.

The gap between using AI and trusting it is now measured yearly. The Stanford AI Index 2026 reports 58% of employees worldwide use AI at work regularly while 53% trust it, and in the same survey 52% say AI products make them nervous (Chapter 9). The skill is what closes that gap, and the prompt is only the first inch of it.

This is exactly what happens when AI replaces thinking rather than extending it. The student who asks AI to construct their argument before working out what they believe has the document, but not the reasoning. The analyst who uses AI to interpret data before forming their own hypothesis has the output but not the judgment. A 2025 global study by the University of Melbourne and KPMG surveying over 48,000 people across 47 countries found that while 66% now use AI regularly, fewer than half actually trust it. The professionals who stand out are the ones whose thinking is present in the output, regardless of how the draft was generated.

The Hierarchy of Competence

The most useful frame for this distinction separates three levels of engagement with any tool. The first level is owning the tool: buying the subscription, knowing it exists. This requires no skill. The second level is operating the tool: knowing how to prompt, understanding what kinds of inputs produce what kinds of outputs. This is what most AI literacy programmes teach. It is useful. It is also the level that will be commoditized most quickly as tools become more intuitive.

The third level is thinking with the tool: using AI to extend and sharpen your own reasoning rather than to replace it. Drafting your own position first, then using AI to stress test it. Forming your own interpretation of data before asking AI to generate alternatives. At this level, the tool amplifies what you bring to it. At the second level, it masks what you do not. The World Economic Forum’s Future of Jobs Report places analytical thinking and creative thinking above technological literacy in its ranking of critical skills. Technology is the floor, not the ceiling.

What Korea’s Own Survey Says People Use AI For

It helps to look at what people in Korea actually do with these tools, because it shows which level most of us are working at. The Korea Media and Communications Commission and KISDI run a yearly panel survey on the intelligent information society, and the 2025 results, published in May 2026, are clear about purpose.

Read the second and fourth cards together. The people using AI mostly use it to find things faster, and the people not using it say they do not know how. Both answers sit at the second level of the hierarchy. The obvious response is more prompting courses, and that would help with the fourth card. It would do nothing for the second, because a faster search is still a search. The question almost nobody is training for is what you do with what you found.

What This Means for How Organizations Train

If your AI literacy training teaches people to prompt, it is developing operational capability. If it teaches people to think and then use AI to extend that thinking, it is developing competitive capability. The first is replicable by anyone with access to the same tools. The second compounds over time. Research published in the Journal of Marketing and Social Research in 2025 found that AI generated content significantly reduces perceived authenticity and brand trust compared to content made by humans. The content that reduces authenticity is not AI assisted content. It is content where human thinking has been removed from the process.

In my technology and AI course, the test is simple: can you explain your reasoning without a screen? If the answer is no, you have not developed the skill. You have borrowed it. The WiFi goes down. The subscription lapses. The tool changes. What remains is what you actually built.

A Worked Example: The Same Prompt, Written After the Thinking

Here is the difference between the second and third levels in a single task. Lee Dohyun is a junior analyst on the merchandising team of a convenience store chain headquartered in Seoul. Lunchbox sales dropped in the third quarter, and his manager wanted an explanation by Friday. His first instinct was the one most training teaches: write a good prompt.

By most training standards the first prompt is the better one. It sets a role, a task and a format. It is also the one that produces an answer Dohyun cannot defend, because none of the reasoning is his. The second prompt is rougher and took him twenty minutes longer, because he had to open the store data and commit to a view first. When his manager asked why the weather was not the cause, he had an answer, and he had it without a screen.

The practical rule I give teams is short. Before you open the tool, write two sentences: what you think is going on, and what would change your mind. Put both into the prompt. If you cannot write the first sentence, you are not ready to prompt yet, and no template will fix that. If you cannot write the second, you are about to ask the tool to agree with you, which it is very good at. Most of the value in the worked example came from those two sentences, not from the phrasing around them.

The research on consultants points the same way. In a field experiment with 758 Boston Consulting Group consultants, summarised by the HBS AI Institute, the AI tool boosted task completion by over 12%, speed by over 25% and performance rated by humans by over 40%, but only for tasks inside what the researchers call a jagged frontier, where AI excels at some tasks and falls short in others. The consultants are described as Centaurs, who divided the work between themselves and the tool, or Cyborgs, who wove it into every step. In both styles, a person was still deciding where the tool belonged. That is the third level.

To try this on your own work, write your view in two sentences, then paste it into Copy Check before the tool ever sees it. If you are not sure which tasks deserve that effort, The Offload Line helps you sort them. The case for why this matters across a whole organisation is in Why AI Literacy Can No Longer Be Optional, the writing habit behind it is in Why the First Draft Is the Most Important Draft You Will Never Show Anyone, and the wider picture of what AI is changing at work is in What No One Tells You About AI in the Workplace.

→  This is one of the topics I speak on for both corporate and academic audiences: the distinction between AI literacy and AI dependency, and what organizations can do to develop the former without producing the latter. The Work With Us page has more on speaking engagements and workshop formats.

If you want practical tools to sharpen how you communicate professionally, the communication tools on this site are a useful starting point.

---