Curry, Depth TechFlow

Recently, a wave of AI unemployment anxiety has swept through the internet in both the East and the West.

Block laid off 4,000 people, with the CEO saying AI can do your job; Pinterest cut 15% of its workforce, redirecting funds to develop AI; Dow Chemical laid off 4,500 people, citing a push for automation...

The domestic scene is no different, with NetEase rumored to be replacing outsourced jobs with AI, iFlytek denying rumors of mass layoffs, and ByteDance reported to be optimizing non-AI departments by 20% every six months...

Statistics show that in the first three months of 2026, layoffs in the global tech industry have exceeded 45,000 people, with nearly 10,000 clearly attributed to AI.

In this context, the Financial Times reported last Friday that OpenAI plans to expand its workforce from 4,500 to 8,000 by the end of the year.

3,500 new positions. Isn't it surprising that a company creating AI says it doesn't have enough personnel?

Upon reviewing OpenAI's recruitment page, it is clear they are hiring engineers and researchers, but densely listed are another type of positions: partner managers, enterprise sales, GTM (go-to-market strategy) teams, and a newly mentioned role called technical ambassadorship, which translates to:

Technical ambassador, specifically helping enterprise clients learn how to use AI.

So, OpenAI is not hiring people to make AI stronger; it is hiring people who can convince others to spend money on AI.

Capturing Clients is More Important than Refining the Model

ChatGPT has 900 million active users weekly, but most of them don’t pay.

For paying consumers, OpenAI is actually operating at a loss: the computational cost consumed by each heavy user exceeds the $20 monthly fee. This year, it is expected to generate $25 billion in revenue and incur a loss of $14 billion.

Consumers support traffic, while enterprise clients support profits. And enterprise clients are flocking to Anthropic's Claude.

Data from Ramp shows that among companies making their first purchase of AI tools, Anthropic has taken 73% of the market share. Ten weeks ago, this figure was more balanced between the two companies.

Last December, Altman sent a "code red" memo to all employees, pausing all non-core projects like advertising and shopping assistants, focusing all company resources on enhancing the ChatGPT experience.

The triggering factor was Google Gemini 3 surpassing ChatGPT in multiple tests, but a deeper anxiety exists on the enterprise side: Anthropic is embedding Claude into clients' codebases and workflows, and once integrated, the cost of migration starts to snowball.

Models can iterate, but once clients leave, they won’t come back on their own. Pursuing clients cannot rely on AI giving suggestions; there needs to be real people knocking on doors.

Shovels Can't Sell Themselves

AI can write code, handle customer service, and conduct data analysis, but there is one thing it cannot do:

Convince a company's tech leader to sign an annual contract to buy from me.

Using AI as an individual is straightforward, just download an app and you can uninstall it anytime if dissatisfied. Using AI as a business is another matter altogether. Data security reviews, internal process transformations, compatibility with existing systems, employee training, any one of these bottlenecks can stall a project.

This is not a problem that can be solved by model scores; it requires someone to sit in the client's conference room to push things forward.

OpenAI clearly understands this. It is not just hiring salespeople; the FT reports it is negotiating with private equity firms like TPG and Brookfield for partnerships to specifically help enterprises implement AI. The essence of this business still requires people to be present.

Block's story is telling the same thing.

Less than three weeks after laying off 4,000 people, the company began recalling some staff. A design engineer was told it was a "mistake," and a tech lead discovered that after his entire team was cut, no one was there to handle key operations, threatening to leave, which led the company to bring back some employees.

Dorsey himself hinted in the layoff announcement: we may have laid off some wrong people...

AI has indeed caused anxiety regarding layoffs, but if the lifeblood is cut because of AI, it is clearly a case of going too far. Even in a company where the CEO openly states AI can replace most employees, there are still components that AI cannot handle.

AI is best at replacing tasks that can be clearly defined, but the task of "convincing an organization that it needs AI and then helping it to use it" just cannot be clearly defined.

Every round of technological revolution sees people saying "selling shovels is the most profitable." This round of AI is no different, with the consensus that companies building infrastructure will profit regardless of who wins or loses.

But OpenAI's current situation indicates that while the shovels have been created, someone still needs to teach others how to use them. Moreover, this "teaching" process cannot be completed using the shovel itself.

Ground Promotion, A Stable Job Amid AI Anxiety

When you look at the people laid off and those being hired, a clear dividing line emerges.

Among the 4,000 people laid off by Block, a significant portion were engineering and operational roles added during the pandemic, performing tasks that can be standardized and described. The 3,500 people hired by OpenAI primarily consist of sales, customer success, and partner management roles, doing work that cannot be documented in workflow documents.

What OpenAI is doing has an old name: ground promotion.

Sending people to the client's office, sitting down, listening to needs, integrating systems, and monitoring launches. Whether it's a technical ambassador or a partner manager, changing the label from English essentially boils down to what Meituan did ten years ago during the O2O battle, sending people door to door to persuade restaurant owners to install POS systems.

This line of thought is seen not only in these two companies.

The CEO of Shopify told employees this year that if they want to hire more people, they must first prove that AI cannot handle the task. Klarna laid off 700 customer service staff two years ago, claiming AI was sufficient, but quietly hired people back last year, with the CEO admitting they had "moved too quickly" on AI.

What is the difference between those laid off and those recalled?

The positions that could be cut share a common characteristic: the work can be broken down into clear inputs and outputs. Writing a piece of code, responding to a ticket, generating a report, all have clear boundaries and AI excels at this.

The characteristics of ground promotion are quite the opposite. Helping a financial client integrate AI into compliance systems and assisting a gaming company in content generation using AI are totally different projects. The people on the other side vary, thus the solutions differ. This task cannot be outlined in a prompt.

AI is not eliminating all jobs; it is re-evaluating the pricing of jobs. Those that can be described in a single statement are becoming cheaper, while those that cannot are becoming more expensive.

Three years ago, a company could change the world with a single paper; now it has to hire thousands of people to knock on doors one by one.

If you are anxious about whether AI will replace you, the answer may not depend on your industry but on whether your job can be clearly articulated in a single statement.

The parts that can be described clearly are becoming less safe.