AI leads to layoffs, but OpenAI is hiring salespeople

Companies developing AI are hiring on a large scale for “ground promotion”—they’ve built the shovel, but still need people to teach others how to dig.

Curry, Deep Tide TechFlow

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

Block laid off 4,000 employees, with the CEO saying AI can do your job; Pinterest cut 15% of its staff, investing the money into AI; Dow Chemical laid off 4,500, citing increased automation…

Domestically, the situation is no different. NetEase was rumored to replace outsourcing with AI, iFlytek denied large-scale layoffs, and ByteDance reportedly restructured 20% of non-AI departments every six months…

Statistics show that in the first three months of 2026, global tech layoffs exceeded 45,000, nearly 10,000 of which were explicitly attributed to AI.

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

3,500 new positions. A company making AI says it doesn’t have enough people?

Check out OpenAI’s careers page—engineers and researchers are obviously being recruited, but what’s equally prominent are other roles: Partner Managers, Enterprise Sales, GTM (Go-To-Market) teams, and a new position mentioned in the report called Technical Ambassadorship, which translates to:

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

So, OpenAI isn’t hiring people to make AI smarter; they’re hiring people to make others willing to pay for AI.

Winning clients is more important than perfecting models

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

Paid users, OpenAI is even losing money serving: the computing power cost per heavy user exceeds the $20 monthly fee. This year, revenue is projected at $25 billion, with an estimated loss of $14 billion.

Consumers drive traffic; enterprise clients drive profit. And enterprise clients are moving toward Anthropic’s Claude.

Ramp data shows that among companies that first purchased AI tools, Anthropic captured 73% of the market share. Ten weeks ago, this was split evenly between two companies.

Last December, Altman sent a “Code Red” memo to all staff, pausing all non-core projects like advertising and shopping assistants, focusing all resources on ChatGPT experience.

The trigger was Google Gemini 3 surpassing ChatGPT in multiple tests, but deeper anxiety existed on the enterprise side: Anthropic was embedding Claude into clients’ codebases and workflows, and once integrated, migration costs would snowball.

Models can be iterated, but once clients leave, they don’t come back on their own. Winning clients requires real effort—AI suggestions alone aren’t enough; someone has to knock on their door.

Shovels can’t sell themselves

AI can write code, handle customer service, analyze data, but one thing it can’t do:

Convince a company’s tech leader to sign an annual contract with me.

For personal AI use, just download an app and uninstall if dissatisfied. But for enterprise AI, it’s a different story. Data security reviews, internal process restructuring, system compatibility, employee training—any bottleneck can halt the project.

This isn’t something a model’s performance score can solve; it requires someone sitting in the client’s conference room to push forward.

OpenAI seems to have realized this. They’re not only hiring salespeople; FT reports they’re negotiating joint ventures with private equity firms like TPG and Brookfield to help enterprises implement AI. The core of this business is still about sending people in.

Block’s story is telling the same thing.

Less than three weeks after laying off 4,000 people, the company started bringing some back. A design engineer was told they were “laid off incorrectly,” and a technical lead found that after the entire team was cut, no one could handle key operations, threatening resignation. Only then did the company rehire some staff.

Dorsey even hinted in the layoff letter: “We may have laid off some people incorrectly…”

AI has indeed caused layoffs anxiety, but if the main arteries of revenue are cut because of AI layoffs, that’s clearly overkill. Even in companies where CEOs publicly claim AI can replace most employees, some tasks still can’t be replaced.

AI is best at replacing tasks that are clearly defined, but “convincing an organization it needs AI and helping it implement” is precisely something that can’t be clearly defined.

Every technological revolution has had people claiming “selling shovels is the most profitable.” This round of AI is no different—most agree that infrastructure companies will be immune to wins and losses, regardless of who wins.

But OpenAI’s current situation shows that while the shovels are made, someone still needs to teach others how to use them. And this “teaching” process can’t be done with shovels alone.

Ground promotion, the iron rice bowl in AI anxiety

Looking at the layoffs and new hires together reveals a dividing line.

Among the 4,000 laid-off at Block, many were from engineering and operations roles expanded during the pandemic—jobs that can be standardized and described. The 3,500 new hires at OpenAI are mainly in sales, customer success, partner management—roles that can’t be easily documented in processes.

What OpenAI is doing is a classic example of ground promotion.

Sending people to clients’ offices, sitting down, listening to needs, integrating systems, monitoring launches. Whether called Technical Ambassadors or Partner Managers, fundamentally it’s like the old O2O battles a decade ago, where Meituan sent people door-to-door convincing restaurant owners to install POS machines.

This approach isn’t limited to these two companies.

This year, Shopify’s CEO told employees that if they want to hire more, they first need to prove AI can’t do the job. Klarna laid off 700 customer service reps two years ago, claiming AI was enough; last year, they quietly rehired some, with the CEO admitting they “moved too fast” on AI.

What’s the difference between those laid off and those rehired?

Jobs that can be cut share a common feature: their tasks can be broken down into clear inputs and outputs. Writing code, replying to tickets, generating reports—boundaries are clear, and AI excels at these.

Ground promotion is the opposite. Helping a financial client integrate AI into compliance systems, or assisting a gaming company with content generation—no two projects are the same. The person on the other side is different, so the solution varies. This can’t be scripted into prompts.

AI isn’t eliminating all jobs; it’s redefining their value. Tasks that can be clearly described are becoming cheaper; those that can’t are becoming more expensive.

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

If you’re anxious about AI replacing you, the answer may not depend on your industry but on whether your work can be clearly explained in one sentence.

The part that can be explained clearly is already less secure.