Everyone is talking about what AI can do. Almost nobody is talking about what it is doing to the people being asked to use it.

Most organizations have now issued something on AI. A policy, a position paper, a set of guidelines, a memo from leadership about the opportunities ahead. What almost none of them have issued is an honest account of what AI adoption actually means for the people doing the work. What it changes about how their judgment is valued. What it means for the skills they spent years building. What happens to their professional identity when a tool can approximate their output in seconds.

This is not an argument against AI. I use AI tools regularly and teach courses that include them. But there is a conversation that is not happening in most workplaces, and its absence is creating exactly the kind of confusion and resentment that organizations claim to want to avoid.

The productivity framing leaves people out

The dominant organizational frame for AI adoption is productivity. AI tools will make your team faster, reduce friction, automate low value tasks, and free up human capacity for higher order work. This frame is not wrong, but it is incomplete in a way that matters. It positions employees as inputs to an optimization process rather than as people who might have legitimate concerns about what their work means and how they are being asked to do it differently.

Korean opinion is unusually clear on which side it wants. In the Stanford AI Index 2026, 70% of Korean respondents said the priority should be advancing science, medicine and other fields rather than protecting industries through regulation, against a global 58% (Chapter 9, Figure 9.3.3). That is the room this conversation has to be had in: not sceptical of AI, but silent about its ethics.

When you tell someone that a tool will handle their low value tasks, you are implicitly telling them that you have already decided which parts of their job have low value. That is a significant claim that most organizations are making without discussion, and workers notice. They may not say so directly, especially in hierarchical environments where dissent is costly. But they notice.

The accountability gap

When AI assisted work goes wrong… and it does, regularly… organizations are still figuring out who is responsible. The person who used the tool? The team that selected it? The vendor who built it? The manager who set the deadline that made careful review impossible? This accountability gap is not a technical problem. It is a communication and governance problem, and most organizations are treating it like something that will sort itself out once the tools mature. It will not.

Clear accountability requires clear communication about what the tool is doing, where human judgment is still required, and what constitutes an error versus an acceptable limitation. Most AI rollouts skip this entirely. The result is that individual workers absorb the accountability that should be distributed across the system, because they are the ones who put their name on the output.

What the silence is communicating

Organizations that are not having this conversation are still communicating something through their silence. They are communicating that employee concerns about AI adoption are not worth addressing directly. That efficiency matters more than how people experience their work. That the decision has already been made and the only thing left to discuss is implementation. People read that clearly, even when nothing is said.

The organizations that are getting AI adoption right are not the ones with the best tools or the most aggressive rollout timelines. They are the ones treating it as a communication challenge first. They are being specific about what the tools will and will not do. They are naming the tradeoffs honestly. They are asking the people doing the work what they are worried about, and then actually responding to the answers.

A Worked Example: The Report Nobody Owned

Jaehyun is a 38 year old team lead at a mid sized logistics company in Incheon. In spring his division rolled out an AI tool that drafts the weekly customs summary for overseas clients. The memo from headquarters was two paragraphs long and mentioned productivity four times. In week six, a summary went to a client in Rotterdam with the wrong tariff code for a container of auto parts. The tool had pulled an outdated table. The junior analyst who sent it had trusted the draft because the deadline had moved up a day. When the client called, everyone looked at the analyst.

Read the left column again. Every line ends with the most junior person in the chain, which is exactly what it looks like when individual workers absorb accountability that should be spread across the system. Nothing in that column is malicious. It is simply what happens when nobody says out loud who owns what. The right column did not need new software. It needed one meeting and a shared document.

In a Korean team this pattern is harder to break, because the person at the bottom of the chain is the least likely to say in a meeting that the deadline made checking impossible. Our guide to how Korean workplace hierarchy actually works explains why. So the conversation has to be opened from above, and soon, because the habit is already forming. In the University of Melbourne and KPMG global study of more than 48,000 people across 47 countries, many said they rely on AI output without evaluating accuracy (66%) and are making mistakes in their work due to AI (56%).

What the Korean AI Basic Act Changes, and What It Does Not

Since 22 January 2026, Korea has had a law that touches this directly. According to the Ministry of Science and ICT announcement on 정책브리핑, a business offering a product or service that uses high impact or generative AI must tell users in advance that AI is involved. The same announcement says that where a person intervenes in the final decision, the system is treated as controllable and falls outside the high impact category, and that the government will hold off on regulation for at least a year so companies can prepare.

Read that second point carefully. The law treats a human in the loop as the safeguard. In Jaehyun’s team, the human in the loop was a junior analyst with no time and no authority to question the draft. A person whose name is on the output but who cannot realistically say no is not oversight. That is the conversation the law cannot have for you, and it is why the questions below matter more now, not less.

There is a practical first step that costs nothing. Before the next rollout memo goes out, add one paragraph to it that names who checks the output, how long they have to do it, and who answers the client when something slips through. If that paragraph is hard to write, the rollout is not ready, whatever the tool can do.

The questions worth asking

If you are in a leadership position, the conversation worth having with your team is not “here is how we are going to use AI.” It is: what are you worried about losing? What parts of your work do you think require human judgment and why? Where do you want a human in the loop and where would automation actually help you? What would you need to see to feel confident about how accountability works here?

Run it with the most senior person speaking last, not first. In my experience that one change does more for honesty in a Korean meeting than any wording of the questions. If you want to check how your own values hold once deadline pressure arrives, The Long Game is built for that. For how small shortcuts like the one in Jaehyun’s team grow when nobody names them, see the escalation ladder, and for the incentive side of the same problem, read who benefits when AI gets it wrong and the AI governance gap.

These are not questions that slow down adoption. They are questions that make adoption actually work. Because tools that people do not trust, or that people feel were imposed on them without conversation, do not get used well. They get used minimally and resentfully, which is the worst outcome for everyone.

The AI conversation your organization is not having is a communication problem. And like most communication problems, the solution starts with deciding that what the other people think actually matters.

CareerComms teaches and advises on technology, AI and communication at a Seoul university and with corporate teams. Work with us here.

---