Throw the boring work to the large model first?

Delegating the uninteresting and boring tasks we don't want to do to large models might be the beginning of AI reshaping work processes and organizational models.

Recently, we often hear the idea of entrusting "boring" jobs to AI.

For example, Greg Brockman, the co - founder and president of OpenAI, also mentioned in an interview at the AI Engineer Summit that AI can handle the migration and update tasks that humans find boring, such as transforming large legacy codebases (Morgan Stanley has been using its self - developed AI tools to convert old code in batches this year).

OpenAI understands users' psychology because this is what workers around the world expect from AI.

Not long ago, a research team from Stanford conducted a study, inviting 1,500 front - line workers from 104 occupations to choose what jobs they would like to delegate to AI.

Ultimately, the top five jobs were: arranging customer appointments, organizing emergency files, correcting payroll records, data format conversion and import, and website data backup.

These jobs have common characteristics: high standardization, frequent repetition, low judgment intensity, but they are extremely time - consuming and error - prone.

In the workplace, we definitely want to hand over these repetitive jobs with low unit - time output to AI, and it seems to be a win - win situation for both us and AI.

However, this is far from the end. From a practical application perspective, it is more likely to be the beginning of AI reshaping work processes and organizational models.

Let AI Do What It Can

The Stanford task list also includes "advanced options" such as content generation, code writing, and creative design, but few people choose them.

These are the things large models did when we first got excited about AI creation. We asked large models to write poems, songs, and draw pictures... Soon, we realized something was wrong, "The large model is writing poems while I'm doing the dishes."

Boring jobs not only bore us but also make us think they are not difficult. However, we overlooked one point: the final results of the same job done by AI and humans may be completely different.

Recently, I saw a post on Xiaohongshu. A technical staff member from a large company said that an AI product developed by his team in half a year led to a 1/4 reduction in the 200 - plus sales team of the department.

This might be the work scenario of robots that many people imagine

This product is a voice AI capable of communicating with people, similar to the AI customer service we often encounter. After the boss tried it, he thought this voice AI could be used for telemarketing.

Now, no one in the sales team does telemarketing because it is a commission - only job with a low conversion rate, so no one wants to do it.

However, AI has no time cost and can make calls tirelessly. After a period of testing, the conversion rate was quite good. So, the sales team didn't need as many people.

The original intention of developing this product was not to compete with salespeople, and it wasn't even developed for the sales department.

AI just did the jobs that salespeople didn't want to do, but the final result was not that all salespeople became more relaxed, but that 1/4 of the team was replaced.

It's not hard to imagine that AI is really suitable for doing such boring jobs. It won't have emotional fluctuations because the call is hung up or it is insulted, won't be interrupted by going to the toilet, eating, or sleeping, and won't give up because of low commissions, thus greatly improving the output of this job.

However, is AI really that intelligent now? In the comment section of the post at the beginning, someone wondered why AI telemarketing could result in deals. Generally, don't people hang up such calls immediately?

The fact is that after testing, AI can conduct several rounds of conversations without being detected. In fact, I also received an anti - fraud call and talked to the caller for a long time before suspecting it was an AI customer service.

There are many such jobs that humans don't like to do. A friend working in a publishing company said that, for example, the theme of a book cover is usually determined by the editor - in - charge based on the content, and then the graphic designer comes up with a design plan.

This process is similar to the design of magazine covers in the past. It requires comprehensive consideration of the theme to be reflected in the content, design style, graphic and text layout style, etc. All these aspects require experience and thinking, and discussions between two teams.

But the sketching part only requires the graphic designer to implement it. Usually, it takes a graphic designer several days to come up with a draft, and when there is a lot of work, it may take more than a week.

Now, as long as the technical team builds a simple workflow and inputs various requirements in the background, ten sketches in different styles can be obtained instantly.

Using AI to improve efficiency has two results: First, it saves manpower. Some editors in publishing companies can even design book covers by themselves without graphic designers. And books with AI - designed covers are already on the market. Some companies still keep graphic designers, but obviously, they don't need as many.

Second, and I think more importantly, the participation of AI has changed the work process. The editor - in - charge no longer needs to discuss with the graphic designer first. Instead, they can take a bunch of sketches and tell the designer what style they want.

The Quiet Infiltration of Large Models

In factories, industrial robots entered the production line by replacing the processes that humans didn't like to do first.

A few years ago, on a refrigerator production line, I saw only one process was replaced by a robot, which was responsible for moving refrigerators in cartons up and down. Originally, this job was done by two workers lifting together, but after an eight - hour workday, the work intensity was extremely high, so it was difficult to arrange manpower during scheduling.

Now, in unmanned refrigerator factories, just looking at the production line, it is basically a lights - out factory, with robots doing all the work in the entire workshop.

This is what a real fully automated automobile production line is usually like

There is an ultimate judgment for the organizational form changed by AI, which is the one - person AI company. Sam Altman, the founder of OpenAI, believes that "the era when one person can build a company worth one billion US dollars is coming soon."

One person, one computer, and an annual income of one million. But this story that was popular last year has basically disappeared on the Internet this year.

It's not that this future can't be realized. It's just that large models can't do many practical jobs now, so there is no suitable business model.

The Agent, which is very popular this year, is exactly to solve this problem. Of course, the capabilities of Agent also depend on large models.

Regarding the capabilities of Agent, in March this year, Fu Sheng, the chairman and CEO of Cheetah Mobile, made a very apt analogy. If measured by the standards of autonomous driving levels L1 - L5, most Agents are currently at the L1 - L2 stage. One day when they reach the L5 stage, people can arrange tasks such as writing documents, researching information, making PPTs, and buying plane tickets for Agent before going to bed at night. When they wake up, they can check the work results.

Actually, in some fields, this goal has already been achieved.

AI musicians have started operating on Spotify. The image and music of AI singers are all generated by AI. It's difficult for users to distinguish between real and AI singers, and real singers even need to prove their identities.

Actually, I've also tried AI - generated music. Using a multimodal large model, I fed an ancient poem into it, and within a few minutes, it generated music in various styles, which actually sounded quite good.

Not to mention AI influencers. Although they are not favored by platforms, they can still earn money by taking advertisements. These digital humans can basically pass for real people. In the comment sections of influencers, questions like "Are you a digital human?" are often seen.

For bosses, the fact that AI can gradually do some work is an undeniable trend.

According to a study released by the Anthropic team in early 2025, employees in 36% of global occupations have used AI for at least a quarter of their daily tasks. A survey by OpenAI also shows that 80% of American workers have at least 10% of their tasks affected by AI, and in nearly one - fifth of the positions, AI has intervened in more than half of the work content.

A friend working on large models said that if your job is related to pictures, audio, or video, you will probably be affected by AI soon.

He said that there are far more companies using large models to improve efficiency than we think, but many bosses are doing it quietly. Usually, they use a small team to first sort out the company's work process and then see which parts can be done by large models.

Once AI enters the work process, the next things may be to change the work process and the organizational model.

This article is from the WeChat official account "20 She" (ID: quancaijing_20she), written by Wang Xiaoling and published by 36Kr with authorization.