Author: Ada, Deep Tide TechFlow

“A major Internet company offered 60 newly graduated PhDs with AI backgrounds a salary of over 3 million this year.” The founder of the headhunting company TTC, which has served over 1,500 AI companies, Xiao Mafeng said this number in a calm tone, as if reporting the weather of the day.

In the same month, data from Maimai showed that the number of AI positions surged 29 times, while Zhaopin reported a 200% increase in job seekers. A 29-fold increase in job openings, a 200% influx; the numbers look beautiful like a bullish K-line.

But this set of numbers hides a secret: a large amount of capital and attention is pouring into a funnel with a very narrow opening. A few dozen people at the top are driving up the entire market's salary expectations, while hundreds of thousands at the bottom are bearing all the anxiety.

And in the middle of the funnel, those who have been in the workplace for five or ten years are being quietly hollowed out.

The prosperity of the talent market is a lie, the illusion of liquidity is real.

One general is hard to find, ten thousand soldiers fight.

The Liepin report shows that 47% of AI positions require master's or doctoral degrees, with nearly half of the companies only recognizing 985/211 graduates.

Headhunter Eva states more directly: “For big companies, 211 is barely acceptable; at least it has to be 985, and resumes without vertical project experience are basically not reviewed.”

What is the peak like?

The day the news of Lin Junyang's resignation from Alibaba Qianwen was announced, “People from major companies all came to us asking if we could help connect with Lin Junyang,” Xiao Mafeng recalled.

There may only be a few dozen talents of this level nationwide. To find them, headhunters have long stopped sifting through resumes. They immerse themselves in GitHub to check code submission records, track paper authors on Google Scholar, and blend into podcast listener groups and AI startup communities. Eva even joined a Tsinghua AI startup competition group, filled with young people in their early twenties. “Chatting early, they may have job needs two to three years later, getting a spot first.”

Another headhunter, Steve, who started recruiting for AI in 2022, made a profound remark: "I seriously doubt there will be resumes in the future."

He gave an example. In January this year, a company wanted to hire someone who understands OpenClaw, a direction so new that no one would write it on their resume. His approach was to break down the demand; it is essentially a multi-agent framework problem. Is there anyone who has worked on similar frameworks? Is there any open-source version? Who are the contributors in the open-source community?

Resumes are devaluing, and traditional recruitment channels are failing.

Some have seized opportunities from this crack.

Sam, co-founder of DINQ, started his business from a similar observation: the top authors of those impressive OpenAI papers often do not come from prestigious universities, some even dropped out, young, with no titles; it's not obvious where their strength lies if you’re not a tech person. The logic of looking at education and experience on LinkedIn fails when applied to AI talent.

Therefore, Sam created DINQ, “LinkedIn for AI scientists and developers,” which does not look at resumes but at achievements: citation counts of key conference papers, contributions to GitHub code, and whether the collaborators are tech elites. If HR inputs "Sora 2" on DINQ, the platform will extend to authors of related technology papers rather than being limited to experience related to Sora 2, digging out the submerged talent.

Xiao Mafeng's alternative solution is to build in public: throw your product out there directly; it’s the best proof of capability.

Although 621 universities have launched undergraduate programs in artificial intelligence, McKinsey predicts a shortage of 4 million AI talents in China by 2030. But the word shortage is misleading; the lack is of experimental scientists who have trained on hundreds of thousands of cards, and a composite talent who understands both the limitations of large models and can identify commercial scenarios. The market has never lacked those who, after listening to two episodes of a podcast, casually say, “I am particularly interested in AI.”

Youke founder Ye Xiangyu summarized it accurately: At the peak, “one general is hard to find,” at the bottom “ten thousand soldiers fight.” Maimai’s statement that “every two AI positions can match one suitable candidate” refers to the peak. How about the bottom? No one is counting because the resumes at the bottom never make it into the system.

Leverage pricing: The closer to the model, the more valuable.

So where is the money flowing to?

Eva provided a set of numbers. For large companies, the ceiling for non-technical positions at level P7 is around 1 million. For equivalent AI technical positions, it ranges from 1.5 million to 2 million. The jump in salary for switching jobs is even bigger, with technical positions commonly increasing by 50%, and some doubling; non-technical positions increase by 10% to 20%, with the maximum not exceeding 30%.

Steve used a term to explain this pricing logic: leverage.

Imagine the model as a sun. The closer someone is to the core, the bigger the lever they can use, and the higher their value. The ability improvements made by a core researcher can affect the market value of a large company by tens of billions. The cost of operating those hundreds of thousands of cards far exceeds their salary. From this perspective, giving them a hundred million is not expensive.

What about those far from the sun? Product managers, operations, sales—the leverage effect is not so direct, and salaries are naturally limited. Steve estimates that at the application layer, the salary gap between technical and non-technical positions is over two to three times.

Xiao Mafeng added a key variable. He believes the essence of this “chain of disdain” is supply and demand, divided into two layers. At a macro level, there are only a few who have trained on hundreds of thousands of cards, and their salaries are sky-high. But at a micro level, it depends on the founding team’s genes. If the founder is a Tsinghua professor with many technical talents in the lab, those who can commercialize become more valuable.

The scarcity of a few dozen people defines the salary narrative of the entire industry. The remaining people use this narrative as a benchmark, which only leads to a gap.

A cleansing of the "old generation."

“In the AI era, we refuse the old generation,” Xiao Mafeng provided a sharp commentary.

Those who experienced the last wave of AI, from Megvii and SenseTime eras, are now probably in their forties, and their experiences have become a burden.

Steve’s statement is more euphemistic but points in the same direction: “We don’t believe that an old map can find a new continent. Those who have been in an industry for too long will have great momentum and inertia. The brain's direct response is the result of strengthened training, but times have changed, and the correct response may be completely opposite.”

Age anxiety has permeated every level. Some investment institutions are searching for entrepreneurs born after 2000, and phrases like “People born after ’95 are already old” are starting to appear.

This sounds absurd, but the signals given by the recruitment market are very real: when resources are limited, the scale tips unhesitatingly towards the young.

“Now the competition is about execution and implementation speed; everyone is training special forces, not large formations,” Steve said. Special forces do not need so many commanders.

But there is a contradiction here that no one is willing to confront directly.

It is precisely the industry experience, tacit knowledge, and the pitfalls crossed that allow for the successful landing of AI products and the conversion of technology into commercial value. Steve himself also admits that this tacit knowledge resides in relatively mature individuals. They might not know the exact future paths, but they know which paths surely cannot be taken.

The industry needs the vigor of young people, but it also requires the judgment of veterans. Everyone can say these two sentences, but the flow of money only addresses the first half.

The middle layer has been swallowed.

Three headhunters coincidentally mentioned a change: the management layer is being compressed.

“It’s probably pretty hard to just manage anymore. A lot of things are being disrupted, and the system you built might be overturned tomorrow,” Steve said.

Organizations are becoming extremely flat, no longer requiring the pyramidal reporting structure, but need small teams where everyone can fight. Relying on people to do something is not as effective as depending on an agent to get the job done. Previously, the emphasis was on strong management abilities and managing complex teams, but this is being challenged.

The boundaries between product managers, operations, and front-end and back-end engineers are blurring. An individual can complete a product's MVP using AI.

Chen Lei (pseudonym) worked as a product director at a mid-sized AI company for three years, managing an eight-person team. Earlier this year, the company underwent a reorganization, her team was disbanded, with four people transitioning to work on agent products while two were optimized out. Her own title changed from “Director” to “Senior Product Manager,” reporting to a technical lead five years her junior.

“I wasn’t laid off, but I know that this feels worse than being laid off,” she said. “What you built over three years in this company was erased by just one organizational adjustment. And you can’t complain because they will say, ‘Aren't you still here?’”

This is the most brutal part of this illusion of liquidity. At the top of the funnel, a few geniuses are fiercely competed over at high prices. At the bottom of the funnel, hundreds of thousands of newcomers cannot even enter the door. And in the middle of the funnel, those who have been in the workforce for five, ten, or even fifteen years are being hollowed out from within.

The career ladder has had its middle few rungs removed. It used to be an elevator, going up one floor at a time; now it has become skydiving, either landing directly at the top or free-falling.

Who is creating this illusion?

In this illusion of liquidity, who are the beneficiaries?

Recruitment platforms use "AI positions have surged 29 times" and "talent gap of 4 million" to harvest traffic, each share pushes more anxious job seekers into the funnel.

Companies use AI as a cover, and Forrester Research found that 55% of employers regret laying off due to AI, as those replaced AI capabilities were not ready. Resume.org’s survey is more direct: 59% of companies admit to wrapping layoffs as “AI-driven,” as it sounds better when explaining to stakeholders. Saying it is because of AI sounds like a strategic upgrade, while saying it is due to poor performance sounds like management failure. AI has become the best cover.

Klarna laid off 700 people claiming AI replaced customer service, resulting in a drop in service quality, customer mutiny, and quietly rehired them. This is not an isolated case. Forrester predicts that half of AI layoffs will eventually be quietly re-hired, but at lower salaries or outsourced overseas.

Steve precisely summarizes the current mentality of bosses: “They first ask the first question, should we hire? Only then do they consider what to hire.”

According to Forrester Research, worldwide, only 16% of employees possess high AI readiness. Companies do not invest in training, relying on employees to learn on their own. Generation Z has the highest AI readiness at 22%, but is the first to be pushed out of entry-level positions, as entry-level positions are the easiest for AI to consume. Mercer’s survey shows that anxiety over job loss due to AI has surged from 28% in 2024 to 40% in 2026.

AI is both the reason for poaching and the excuse for layoffs. Whoever holds the defining power is the house in this game.

The funnel will not widen.

Returning to the initial set of numbers.

A 29-fold increase in job openings, 200% job seeker influx, 3 million annual salary, 4 million talent gap. Each number is real, but when pieced together, they tell a completely different story: jobs are increasing, but the opening is extremely narrow; job seekers are increasing, yet the vast majority cannot even pass screening; salaries are soaring, but only belong to a few dozen at the peak; the gap is widening, yet what is missing does not match what is supplied.

But this funnel will not widen. AI technology iterates every six months; today’s hottest direction may become obsolete in half a year. You might think you are close to the sun; a new model release could throw you to the periphery.

Steve said a phrase that could serve as both a tombstone for this industry and a ticket for entry: “Measuring qualification by the length of time may no longer be sufficient. What matters is the density and depth of your interaction with AI. Someone entered the field four years ago but only used it generally. Someone joined last year but is fully invested. You tell me, who has deeper qualifications?”

The three headhunters are also being reshaped by this industry. Eva is studying algorithm principles, Steve is researching agent frameworks, and Xiao Mafeng just came from a meeting with a post-2000 entrepreneur, lamenting that “their cognition has reached the next level.” Those selling shovels must also keep pace with the rhythm of gold mining.

Chen Lei recently started a small project on GitHub, creating an automated legal document generation tool using an agent framework. No one asked her to do it, and no one is paying her. She said she realized one thing: rather than waiting to be filtered by the funnel, it’s better to carve out a hole for oneself.

This may be the only somewhat optimistic part of the entire text, but it is still just close.

The vast majority of people are neither one of the 60 receiving 3 million nor are they individuals like Chen Lei who still have the capability and willingness to carve a hole. They are the silent majority in the middle of the funnel, not top-notch enough to be fiercely competed over at sky-high prices, yet not resolute enough to start over.

This funnel will not widen.