Wage earners start "performing" AI-driven efficiency improvement.

With just over two months left in 2026, employees at major tech companies are about to get through a whole year dominated by AI.

At the start of the year, Jensen Huang's remark that "if an engineer earning $500,000 a year consumes less than $250,000 worth of Tokens annually, it would make me deeply uneasy" sparked the Tokenmaxxing wave. Everyone was using AI and burning through Tokens, as if using too few Tokens meant their work was not done properly.

By June, major companies realized they could no longer afford the massive Token consumption, and began to shift their focus to the return on AI usage. This did not reduce the AI FOMO (fear of missing out) among their employees, but instead made it harder for them to "please" their bosses by using AI. They had to prove that AI was helping improve their work efficiency, ensure their deliverables had sufficient "AI content", and demonstrate that every Token they consumed was generating value.

For some employees at large companies, whether AI can truly improve efficiency is secondary. What matters more is to show their teams and managers that "I am already using it". For them, it does not matter if you cannot use AI, what matters is not letting your boss think you cannot.

To this end, some people rewrote "having AI transcribe the boss's 30-minute meeting recording into text" as "building a structured process for voice information", which was even rated as an excellent AI application case; some deliberately picked unimportant work scenarios to develop Skills, then packaged them under a fancy-sounding name to show their bosses; some PMs (product managers) at large companies used AI to do the work of RDs (research and development engineers), and then asked their RD colleagues to rewrite the entire work afterwards.

This phenomenon is becoming increasingly common in workplaces at home and abroad. A recent survey involving business intelligence firm Visier showed that among 1,000 surveyed full-time employees in the US, 48% admitted to exaggerating their AI usage frequency or relevant expertise to colleagues or superiors.

It is predictable that as AI continues to deepen and popularize in various work scenarios, this kind of "performance" will continue.

01. To add AI content to my weekly report, I spent half an hour on a 10-minute task

Li Zhao | Born in 1992, Product Operations at an e-commerce company in Hangzhou

I work as a product operations specialist at an e-commerce company in Hangzhou. This March, our OKR required us to achieve "in-depth AI application". At bi-monthly meetings, leaders asked each person what problems they had solved with AI in the previous week. Those who could not give an answer would receive a private message from their supervisor after the meeting.

From March to May, the competition was extremely fierce. There was a usage ranking in our team, and the admin exported the data to the group chat every week. Back then, we did not use AI to get work done, but to avoid falling behind on the ranking. I asked the top-ranked girl how she did it, and she said she asked AI to write the annual plan before going to bed, without even checking the results. I followed her example: as soon as I sat at my workstation, I let AI run "competitor monitoring", and then I would scroll through short videos and read entertainment gossip to get by.

In June, the trend changed: we no longer focused on usage volume, but on output. We needed to present "AI efficiency improvement cases" with data in our weekly reports. I started figuring out how to make up the numbers. The most ridiculous case was that I rewrote "having AI transcribe the boss's 30-minute meeting recording into text" as "building a structured process for voice information", and this entry was later rated as an excellent case. Earlier this month, I drafted a channel price adjustment notice, which would have taken me only 10 minutes. But to add AI content to my weekly report, I let AI draft it first, and it made up a lot of data. In the end, I found last year's Double 11 notice, revised it on that basis and submitted it. But I still wrote "AI-assisted writing" in my weekly report.

Image source / pexels

Making up the numbers is one thing, work still needs to be done. I gradually worked out a set of methods. I have two monitors at my workstation, running AI on the left and working on the right. For plan-related work, I throw the brief into AI and wait for it to generate the framework, without repeatedly adjusting the prompts, and I just answer whatever questions it asks. While it is generating the first part, I write the requirements for the second part; when the second part is running, I go back to check the result of the first part, and finish detailing the third part at the same time, working non-stop for hours. AI does help improve efficiency. Last week, I had more than 40 customer service complaints, and I asked AI to categorize them by issue, and it generated a table in 10 minutes. For sorting out meeting minutes, as soon as everyone got back to their workstations after the meeting, I had already sent the minutes to the group chat.

But AI does not take a fixed time to run tasks, it can take anywhere from 30 seconds to an hour. I tried to do other things while waiting, but when it finished halfway through my revision, I switched back to check the result, then switched back, and often forgot where I left off. Later I simply scrolled through my phone, watching the AI progress at the same time. During that period, I clearly felt my concentration declining. I switched between screens all the time, and I was exhausted after a whole day, but could not tell what exactly made me tired. So I had to categorize my work: when working on task A, I never use AI to do task B. I save small polishing tasks together, and spend one or two hours every day to handle them all at once.

Recently, the department started asking everyone to report their "AI usage insights". I read the summary document the leader sent in the group chat, which said it had integrated everyone's insights. I did not even need to look at it to know that it was all AI-summarized content about AI.

02. The Demo I wrote is completely unusable, but my leader is very satisfied

Li Mo | 27 years old, Product Manager at a major internet company in Beijing

I am a product manager at a major internet company in Beijing, 27 years old, and I have worked here for two years.

At the start of this year, the company began to require all staff to work on AI and write Skills. The R&D colleagues were the first to be affected, and then the requirement spread to all positions. Actually, I had used AI intermittently before, to write copy and sort out requirements, which was at the "trying it out" stage.

Around March or April this year, my team leader talked to me, saying that PMs who cannot write code would soon be eliminated, and asked us all to learn to write code with AI, try to solve simple requirements by ourselves instead of turning to R&D for everything.

After that, I did write some small demos and did some small UI designs. To be honest, none of those things were really usable. When I gave them to R&D colleagues, they had to rewrite everything from scratch, which means no matter how lively my work looked in the beginning, they still had to take full responsibility in the end. But the leader was very satisfied. During that period, I clearly felt that reporting and handing in work had become easier than before. Anyway, whether the product can be implemented is secondary, the posture of "you are using AI" comes first.

Image source / pexels

After June, our team changed a lot, with major adjustments in both front-end and back-end teams, and my reporting line also changed. The new leader has much more detailed assessment standards for AI output and efficiency improvement. The old way of casually writing something and then handing it over to R&D colleagues to revise can no longer pass the assessment.

So I changed my way of "AI efficiency improvement". I wrote all steps including requirement feedback, market research, and product framework into Skills, which are actually very simple page reading and sorting tasks, letting AI handle every link, and the deliverables are basically directly generated by AI. There is not much real valuable content in them, and the "performance" component is more prominent.

For one thing, many of these tasks do not need AI at all, I can finish them in very little time by myself; for another, many of the requirements given by AI are hard for colleagues to push forward, with logical errors often found, so we have to hold meetings to discuss them again. On the surface, the process is running, but the actual efficiency is not necessarily higher than before.

However, this way also has its advantages: the leader thinks we are very busy, the AI workflow is built, the process data looks good, and the reports also look good.

My own judgment on this matter is that AI can indeed improve efficiency, especially in coding and documentation writing, it really saves time. But it cannot always improve efficiency, forcing it into irrelevant tasks will only cause extra trouble. And I personally think the biggest problem is that once the leader thinks your efficiency has improved, he will want you to do other people's work. The dividends of efficiency improvement do not fall on you, but become the basis for assigning more work to you, which is not worth the loss.

I still think that it is a better state for everyone to do what they are good at, and make good use of AI on this basis. Otherwise, if the company continues to do this, it will only spread AI anxiety, waste resources, and finally force everyone to put on a performance.

03. Performance will be deducted if AI usage rate is below 70%, we have rolled out a bunch of useless Skills

Xiao Nan | Post-95s, Event Planner in Shenzhen

I work as an event planner at an enterprise in Shenzhen. Our boss is very enthusiastic about AI, hoping to embed it into every work link, requiring everyone to use AI to assist their work as much as possible, and encouraging us to build our own AI workflows.

Later, the company even included AI usage rate in the performance appraisal: if the usage rate does not reach 70%, 20% of the performance score will be deducted.

At first, everyone thought it was no big deal, after all, AI does help us improve efficiency. It used to take at least 3 days to make an event planning scheme. Now with AI, we can finish the first draft within a day. AI can also help with trivial tasks such as competitor analysis and PPT production, which really saves a lot of time.

But gradually, things started to go wrong. The company held an internal AI competition, comparing who can use AI to cover more work scenarios, build more workflows, and make more Skills. As long as you can prove that you have used AI to open up a certain work link, it can be counted as a work achievement.

At the beginning, the things everyone made were really useful, for example, the Skills related to event planning can reduce some repetitive trivial work.

But the number of valuable scenarios is limited. To complete the tasks, everyone started to compete for quantity. Some people deliberately find unimportant work scenarios to make into Skills, and then package them under a fancy-sounding name; some use AI to make a small program, and show in the report what problems they have solved and how much time they have saved.

Image source / pexels

The boss saw so many new tools and thought everyone was very enthusiastic about using AI. But in fact, some of these tools are of little help to work, and no one uses them after they are made.

I feel that everyone has gradually formed a subtle tacit understanding. Compared with whether AI is really helpful, it is more important to be able to show visible achievements. After all, how many Skills you have made and how many work scenarios you have covered can be easily "faked", but how much time you have saved and how many practical problems you have solved are not that easy to measure.

In addition to these AI achievements specially made for demonstration, a phenomenon has also appeared in daily work: AI is creating new workload.

A colleague of mine quickly generated an event plan with AI, paired it with a beautiful PPT, and reported it to the boss. The boss was very satisfied after reading it, thinking the efficiency was particularly high, and immediately asked everyone to push forward according to the plan. But in the execution stage, we found that there were many loopholes in the plan, and many practical problems were not taken into account. In the end, we still had to rely on people to supplement and adjust it bit by bit.

What makes me more helpless is that the existence of AI also increases the number of plan revisions. In the past, after a planning scheme was finished, the leader would read it once, put forward some opinions, and we could push it forward after revision. Now after the leader reads it, he will send it to AI for review. AI often puts forward a lot of revision opinions, some of which are really valuable, but some are not in line with the actual situation, so we have to check, explain and revise them one by one.

I feel very contradictory. The boss hopes AI can save time for the company, but we have to spend extra time participating in competitions to prove that we are really using AI. With the help of AI, we finally improve the efficiency, but then we have to take on more review, revision and delivery tasks. In this case, using AI has become an uneconomical deal.

04. We are performing for AI: speak with perfect pronunciation, and make your smile audible

Wang Min | Post-95s, Sales at a software development company in Shanxi

I have worked at this company for 5 months, and I have seen what "AI acclimatization" really means.

The company's main business is software development, it has been established for more than ten years, with about 100 employees. The boss is in his sixties, and his understanding of AI basically comes from marketing accounts. Last year, he proposed AI transformation, and at the beginning of this year, it was directly upgraded to the annual strategy. After more than half a year, the output of some businesses doubled after using AI, but more often, the results are ridiculous.

First of all, AI participation was added to the employee performance appraisal, and the performance evaluation depends on the proportion of AI usage.

So everyone started to show their special skills to "pad the AI usage data". For example, I made a skill the day before yesterday, did a data cleaning yesterday, and made a full life cycle plan for the customer from scratch today, the AI usage rate has been rolled up from 30% to 80%. It does not matter whether AI actually works, as long as it "participated".

In the past, weekly reports were written for the boss, but now it has become a circular process: I send my ideas to AI, AI writes the weekly report, I check it and send it to the boss, the boss uses AI to raise a few questions and send them to me, I send them back to AI, and so on. The thing I use AI to do the most every week is to let AI make up what I did with AI this week.