AI erases the traces of labor, and also turns honesty into a kind of workplace risk.

I bought a case of beer on Taobao, and casually asked the customer service rep a question: The alcohol content marked on the bottle is 4.5%vol, does that mean the usual 4.5 degrees? When she typed back, my heart skipped a beat. She said: "I just asked Doubao, and yes, if you're in a hurry later, you can ask it yourself, it's much faster than me checking the backend."

If she had only replied with a simple "yes", I probably wouldn't have thought much about it. But because she added the extra line "I just asked Doubao", I first doubted her professionalism: she's selling the store's own products, why would she even need to ask an AI about this?

But after thinking it through carefully, she was right, and her answer was faster than if I had exited Taobao and searched for confirmation on my own. She didn't make up an answer, didn't delay, and didn't pretend she already knew the answer. What really made me uncomfortable might not be whether her answer was correct, but that she answered so honestly that I first doubted whether the answer came too effortlessly.

At the end of the day, what might need to be questioned is not whether she put her heart into selling the products, but our old way of judging "dedication". This old method should have been replaced long ago.

The same results, honest people get penalized first

In June this year, Teamwork Lab under Atlassian conducted a controlled experiment with nearly 1,000 participants: for the exact same work output, as long as a note was added that said "this was completed with AI help", the evaluators would rate the person's laziness ten times higher, and the proportion of people willing to recommend this person for an important project dropped by 24 percentage points, even though the deliverables submitted by the two groups were actually identical. The only difference was whether they admitted to using AI.

What's more contradictory is that these people no longer need to hide their AI usage: 94% of American office workers used AI last month, and about three-quarters of them are willing to openly tell their colleagues and bosses. Two years ago, Slack conducted a survey of 17,372 office workers, and back then, 48% of people were embarrassed to let their bosses know if they used AI at work, for reasons like it felt like cheating, made them look incompetent, or made them look lazy. This number is no longer accurate today, not because no one uses AI, but because far more people are using AI openly than those who hide it. But even with more people using AI openly, the associated penalties haven't disappeared.

Telling the truth was supposed to earn more trust, but this time, people who tell the truth are the first to suffer losses.

Efficiency goes to the company, but the risk falls on the individual

The result is already there, but the visible effort that used to go with it has been taken away. The result might be completely correct, but all the process traces that used to prove diligence, competence, and sincerity are gone, and trust hasn't caught up yet. We might as well call this gap a "process deficit". This isn't an established academic term, just a temporary name to make it easier to explain this situation clearly.

Few people will take the initiative to fill this gap, because filling it doesn't bring direct benefits to anyone. That day, when she used AI to answer faster, I, the buyer, benefited by saving a few minutes; the store didn't lose out either, since the problem was still solved; the only one bearing the risk was her alone — answering too quickly made me doubt her professionalism in the first place. On a larger scale, the same math applies in companies: employees deliver faster with AI, the company gets higher output, and customers get faster responses. But once this efficiency is suspected of being "watered down", it's never the organization that takes responsibility for it, but the specific individual who handled the task. Efficiency belongs to the organization, convenience goes to the customer, and the risk, along with the suspicion of "being lazy", falls to the individual.

If you can't judge the result, just see how tired they look

But why was "visible effort" so effective in the past? Psychologist Justin Kruger and his colleagues verified this back in 2004 with a rather mischievous method. They showed the same poem and the same painting to two different groups of evaluators, and only told one group that "the author polished this work repeatedly for a long time". That group gave higher quality scores and was willing to pay more for the work. The work itself didn't change, only one sentence was added. This finding later got a name: the "effort heuristic". Subsequent replication experiments haven't produced completely stable results, some support it while others failed to reproduce it, so it's not an absolute iron law, but the core direction it points to still largely holds: the harder it is to directly judge the quality of a task, the more likely people are to use visible hard work as a substitute evidence. Verifying whether an answer is correct requires expertise and takes time; judging whether a person looks like they're working hard only takes a quick glance: if you can't tell whether a proposal is good or not, just count how many versions they revised; if you don't know how big a person's contribution is, just see how late their workstation light stays on, how many pages their weekly report is, and whether they're still replying in the work group late at night. AI just erases all these clues: it polishes answers too fast and too cleanly, so fast that no trace of struggle can be seen.

This kind of psychology doesn't only appear when asking customer service questions. A particularly well-crafted apology text message, or a well-prepared explanation — if it reads too smooth and too complete, it's more likely to make people wonder: did he really come up with this himself, or did a tool think it out for him? The more skilled a person is at writing, the more likely they are to run into this kind of suspicion. If the writing is poor, everyone will think the person wrote it themselves; if it's too clean, so clean that no trace of clumsy struggle can be seen, suspicion will creep in. Clumsiness used to be sincerity itself.

The time saved by AI will never be returned to you

In the past, if a proposal was delivered in two days, the boss would think the work really does take two days; but now if you use AI to deliver it in two hours, he's more likely to think not that your ability has improved, but that you can finish three more proposals in the afternoon.

After this old method stops working, some companies don't rush to establish new evaluation methods, but continue to make employees prove "I really put in the effort": write more detailed weekly reports, specify which work used AI, how long it would have taken originally, and how much time was saved now. But once a person proves their efficiency has improved, what awaits them is rarely leaving work early, but more tasks, until the saved time is filled up again. What AI improves is not the free time that employees can dispose of, but the basis for organizations to re-estimate workloads: if a person compresses a full day's work into two hours, it won't be long before two hours becomes the normal working time for that task. The efficiency created by individuals will quickly become the new benchmark set by the organization.

The logic behind this isn't complicated: verifying "how good is the AI-assisted output" requires professional judgment and has a high cost; judging whether a person looks busy only requires looking at their task volume, attendance, and online status, which is far less costly. Before new methods are established, companies will naturally continue to use the cheapest old indicators. Employees quickly figure out this rule too: since being honest about your real efficiency might bring more tasks, the most rational choice is no longer to be candid, but to keep putting on a show.

Some people might say that as long as the result is good, no one really cares about the process. But in reality, promotions, salary raises, and customer satisfaction ratings are never as simple as just looking at results — they always contain a vague, intangible sense of "putting in the effort", and that feeling is exactly the first thing AI erases.

One person being honest is not enough

Atlassian's experiment also tested a remedial method. When admitting to using AI, describing it as "to help the team" instead of "to save myself time" made others rate your effort 11 percentage points higher, and the probability of being recommended also increased by 8 percentage points — it's a clever wording trick. But no matter how sophisticated the wording is, it's just picking the best option from a bad bunch: these people's scores are still lower than those who "didn't mention AI at all". What can truly make the penalty disappear is never clever wording, but the workplace culture: leaders take the lead in using AI, colleagues treat AI as a shared tool on the table rather than a petty trick someone hides, and at that point, people who use AI a lot will actually get higher scores than those who hide their usage. At the end of the day, one person daring to be open can't hold up against this old system; we need a group of people to come together and establish new rules.

What really needs to change is not how employees explain their use of AI, but whether companies are willing to shift their evaluation criteria from "how much time did you spend" to "what judgments did you make, what problems did you solve, and which results are you willing to take responsibility for".

Later, I asked that customer service rep another question. She answered even faster, and it was that same line again: "I asked Doubao". This time I didn't second-guess it. At least in this store, using AI is no longer something that needs to be hidden. AI hasn't eliminated laziness, it just made us lose the old evidence for judging who's being lazy. We haven't learned yet how to trust a person who no longer "breaks a sweat" at work.

This article is from the WeChat Official Account "Outside Singularity", author: wiwi, published with authorization from 36Kr.