After adopting AI, the company has stopped recruiting new employees.

Today, I'd like to talk to you about something that I think will become more and more common in the future.

A friend's company has reduced its recruitment plan by about half this year, but the recruitment budget remains the same. All the money is used to configure Agents and purchase Tokens.

Simply put, they are using the money originally intended for hiring people to build their own AI employee teams.

As far as I know, they are currently mainly investing in AI employees in customer service, content production, and business planning positions, and the results are quite good.

Note that during this process, the company's expenditure has not decreased, but the number of people has decreased, which leads to a reduction in the organization's "entropy increase".

In plain language, the fewer people and processes, the higher the work efficiency.

I think the reason they can use AI to replace some existing positions still depends on the previously accumulated SOPs and data.

For example, in content production and business planning, they have accumulated SOPs for every case they've handled in the past, and they've also saved some excellent cases as data.

Currently, they are using a combination of Claude Code + OpenClaw, with a hybrid architecture for the models.

For high - demand tasks, they use top - tier foreign models like Claude Opus and GPT. For some basic tasks, they use some cheaper domestic models.

There is a premise that is easily overlooked here: Positions that can be taken over by AI must be those where the workflow has been clearly defined before.

If a position doesn't even have an SOP and people themselves can't clearly explain how to do the job, AI definitely can't do it either. This is actually what I mentioned before as "workflow definition".

The reason I want to talk to you about this is that it's not just a recruitment trend, but there are two completely different issues hidden behind it.

The first issue: AI has indeed increased the efficiency of many positions, but the quality is not necessarily guaranteed.

A content team used to need 5 people to maintain the daily output rhythm. Now, with 3 people and AI, the output is even greater.

A business planner used to spend a week writing a proposal. Now, they can hand it over to AI and get a first draft in ten minutes, and then spend twenty minutes revising it before submission.

With the increase in efficiency, the boss will naturally think that since 3 people can do the work of 5, why should they hire a fourth person?

As a result, recruitment has been frozen.

It's not about laying off employees, but rather not adding new ones. During this process, the company's human resources budget has not decreased, but the efficiency has indeed improved.

Think about it. If a company has 20 positions and each position has a 30% efficiency increase due to AI, theoretically, the workload of 6 positions has been replaced by AI.

These 6 people may not be laid off immediately, but these 6 positions will not be filled with new recruits.

This is why many people clearly feel that there are fewer job openings, higher requirements, and more intense competition when applying for jobs this year.

It's not that the industry is bad; it's that the efficiency structure has changed.

But there is an even deeper change here: In the past, companies hired people for their "executive ability". Now that AI has leveled the playing field in terms of execution, what companies really lack is "judgment ability".

Using the same AI, some people can achieve in a week what used to take a month, while others find that it's even worse than writing by hand after half a day. The difference lies not in the tool, but in the people using the tool.

The ultimate manifestation of this difference is quality.

So you'll find that companies are not not hiring; they're not hiring people who can only execute.

However, there is another side to this matter.

The second issue: Many companies are using AI in the wrong way.

I've observed a very typical phenomenon. Some companies, in order to embrace AI, have handed over almost all tasks that can be done by AI.

What's the result?

The efficiency seems to have increased, but the quality is declining.

I once asked a friend who works in a media company. Their company now uses AI to generate all their brand promotion content.

The output of content such as articles, pictures, and videos has tripled, but the user interaction rate has dropped by 40% compared to last year.

The boss is happy to see the output figures, but no one cares whether anyone reads the content.

There is a counter - intuitive fact here: When all companies are using AI to mass - produce content, the things generated by AI have become noise. What really touches people is the content with a human touch and real judgment.

A classmate from a product team of a large company told me that their internal leader forced everyone to use AI to improve the workflow, requiring them to use AI to write product documents, write requirement descriptions, and draw prototype diagrams.

The documents are written, and they look well - formatted and logically clear.

But when the developers receive them, they find that many key judgments are vague, the boundary conditions are not well - thought - out, and they need to confirm things back and forth more times than when the documents were written by humans.

See, this is the result of blindly using AI.

To put it simply, AI helps you quickly generate something that seems qualified, but it doesn't help you do real thinking.

If you think carefully, you'll find that the core value of some jobs doesn't lie in the output itself, but in the judgment, trade - offs, and control of details during the output process.

If you ask a senior product manager to write a requirement document, the process of writing is actually making decisions.

Which functions to implement, which not to, how to set the priorities, and how to define the boundaries. These judgments are hidden in the writing process.

If you let AI write it, it can produce a beautiful document, but it can't make those key decisions for you. You think you've saved time, but in fact, you've just postponed the decision - making to the development stage, which is even more costly.

There is also a more hidden risk. Young employees who rely on AI too early seem to have high output efficiency, but their underlying judgment ability has never been trained.

When it's time for them to make independent decisions, you'll find that they can only use AI to generate options and can't make their own choices.

It's like a person who takes the elevator every day. They seem to reach the top floor every day, but their legs can no longer climb the stairs.

Speaking of user feedback, many companies are already using AI for it.

They use AI to analyze user feedback. They throw in thousands of comments, and AI classifies, summarizes, and extracts keywords for you. It seems very efficient.

But an experienced operator can spend two hours going through those comments and read the emotions, scenarios, and unspoken needs of users between the lines.

This kind of insight is currently beyond the reach of AI.

AI can tell you what users have said, but it's difficult for it to tell you what users haven't said. And the real problems or opportunities often lie in what users haven't said.

So in reality, in some scenarios, using AI is indeed fast, while in some scenarios, using humans is more accurate.

Where is the problem?

The problem is that many companies haven't distinguished which tasks should be done by AI and which must be done by humans. They regard AI as a "universal substitute" rather than an "efficiency lever".

In my opinion, these two positions are very different.

The logic of a substitute is to use AI for everything that can be done by it, reducing the number of people and costs.

The logic of a lever is that humans make judgments and decisions, and AI does the execution and acceleration. Humans' time is freed up to do more valuable things.

The former saves money in the short term, but in the long run, you'll find that the quality is declining, the decision - making is becoming vague, and the core capabilities are being lost. The latter is the real way AI should be used.

I also have a very clear principle when using AI myself. For anything that requires my personal judgment, AI can only assist, not replace.

For tasks at the execution level, with high repetition and low judgment requirements, AI takes over completely.

The boundary between these two is different for each person, each position, and each business, and you need to figure it out for yourself.

So, going back to the phenomenon mentioned at the beginning: After using AI, companies stop hiring.

This in itself is not a bad thing. Efficiency improvement will change the human resources structure.

But if a company stops hiring because "AI makes us not need people" instead of "AI makes everyone stronger", I think this company will have problems sooner or later.

AI is a productivity tool, not a thinking tool.

What can be completely replaced by AI are only those execution actions that don't require thinking. And the really valuable jobs are precisely those judgments, insights, and creations that AI can't do.

Don't use AI blindly, and don't be blindly afraid of AI.

Since this is an irreversible process, what we need to do is to go with the flow.

This article is from the WeChat official account "Tang Ren" (ID: RyanTang007). The author is Tang Ren, and it is published by 36Kr with authorization.