Who is in demand and who is being phased out?

As the year draws to a close, it's once again the time for major companies to flex their muscles by offering generous year-end bonuses. This year, JD.com and ByteDance have significantly increased their year-end bonus allocations; JD.com's investment in year-end bonuses has surged by over 70% year-on-year, while ByteDance has officially announced a 35% increase in bonuses and a 1.5-fold increase in salary adjustment budgets.

Not long ago, BYD and CATL also announced consecutive salary increases. The competition for talent among leading companies has escalated to a strategic level, particularly for talent in the AI field. According to Maimai's <2025 Annual Talent Migration Report>, from January to October 2025, the number of newly released AI-related positions soared by 543%, with monthly year-on-year increases reaching up to 11 times, far surpassing other industries. Positions in search algorithms and cloud computing are even in a state where multiple positions compete for a single job seeker.

However, not all AI practitioners are highly sought-after. As the large language model sector enters its third year of explosive growth, the AI field is gradually experiencing an oversupply of talent. At the lower end of the pyramid, a significant number of practitioners still face the harsh realities of job contraction and unclear career paths.

01. Major companies 'poach' top talents

Recently, Manus, born out of Zhongguancun, was acquired by Meta (formerly Facebook's parent company) for billions of dollars. Xiao Hong, the founder of Manus born in 1993, became Meta's vice president, causing a stir in the domestic tech circle. Just two weeks before Xiao Hong's promotion, Tencent announced that Yao Shunyu, a 27-year-old former OpenAI researcher, would serve as the chief AI scientist in the "CEO/President's Office," directly reporting to Tencent President Martin Lau, which also sparked industry-wide discussions.

Tech giants both domestically and internationally are hunting for top AI talents with unprecedented intensity. High salaries and acquisition prices have become commonplace in the talent war.

In the first half of this year, Meta poached at least 16 top scientists or engineers from companies such as Anthropic, Apple, and OpenAI, driving up the salary levels for AI researchers in the industry.

Domestically, major companies are also poaching talents from each other with high salaries. Since last year, ByteDance has successively poached Zhou Chang, the former technical leader of Alibaba's Tongyi large model technology team, and Wu Yonghui, the former vice president of research at Google DeepMind. ByteDance's "Top Seed" recruitment program launched in 2024 also targets a large number of PhDs from top universities in large models, AI for Science, and other research directions, as well as authors of highly influential papers at top conferences.

To retain its internal technical talents, ByteDance recently implemented additional stock option grants for core Seed employees. Based on performance and job levels, employees can receive ByteDance stock options worth approximately 90,000, 110,000, or 130,000 RMB per month, with an estimated maximum payout of up to one million RMB over 18 months. This move has even caused dissatisfaction among employees in other departments of ByteDance.

However, ByteDance's core Seed team is also facing poaching attempts from other major companies. "Qujie Business" noticed that in 2025, seven key research personnel have left Seed. Among them, Qiao Siyuan from the large language model team joined Meta; Jiang Lu from the visual model research team joined Apple; and Feng Jiashi, the head of the visual basic research team, joined Tencent as the head of the multimodal team at the AGI Research Center.

Tencent's offensive in snatching AI talents has also significantly increased this year. According to "36Kr," in the past few months, Tencent has been actively poaching top AI talents from ByteDance with doubled salaries; during campus recruitment, Tencent can offer up to twice the salary to poach desired candidates.

Alibaba and JD.com have also launched recruitment and training programs for top AI talents this year, namely the "Alibaba Star Top Talent Recruitment and Training Program" and the "Top Young Technical Genius Program." Under the fierce competition among companies, the salary levels for AI research positions have been on the rise.

An investor revealed that some computer science PhDs from top universities have been offered annual salaries ranging from 3 to 4 million RMB by some major domestic companies. Data from the Maimai report shows that the daily salaries for many AI-related interns in some companies have reached as high as 4,000 RMB, far exceeding the monthly salaries of formal employees in many industries.

According to the Maimai report, from January to October 2025, AI positions dominated the top 20 high-paying positions, with AI scientists/leaders topping the list with an average monthly salary of 127,225 RMB, followed by large model algorithms (71,060 RMB) and digital front-end engineers (69,197 RMB). These high-paying positions mostly come from major companies. The report also mentions that the companies with the highest number of newly released AI positions in China this year are ByteDance, Xiaohongshu, and Ant Group.

The influx of talents has also led to unprecedented mobility among top practitioners in this field, with Resignation and entrepreneurship (leaving to start a business), lightning-fast job hopping, and entire teams leaving becoming commonplace. In the first half of this year, four senior executives from Zhipu AI, including the vice president and chief strategy officer, resigned within a month.

Meanwhile, many companies have also reorganized their structures with AI business as the core. In November, Baidu established two new model research and development departments within its technical platform group (TPG) responsible for the development of the Wenxin large model: the Basic Model Research and Development Department and the Applied Model Research and Development Department, both reporting directly to Robin Li. Tencent also upgraded its large model research and development structure in December, establishing three new departments: the AI Infra Department, the AI Data Department, and the Data Computing Platform Department.

The division of talents and the increased importance of departments indicate that AI is officially transitioning from a "technical support" role to the core engine of corporate strategy.

02. Whose 'rice bowls' are being taken by AI?

Besides the enviable benefits, AI-related teams also stand out in terms of "overtime hours." Media reports indicate that artificial intelligence researchers in Silicon Valley worked between 80 to 100 hours per week in 2025.

Domestically, Yuanbao, Doubao, and Qianwen have also significantly increased their iteration speeds. A former Tencent Yuanbao employee from Guangdong stated on social media that they were almost releasing a new version every three days, practically living at the company. An employee responsible for model algorithms at ByteDance mentioned on social media platforms that they usually arrive at 11 a.m. and can't leave until after 10:30 p.m., discussing work with colleagues during meals, and feeling "overwhelmed by the competition."

Frequent overtime and increased pressure are not limited to top talents. As more job seekers enter the AI industry and the requirements for related skills gradually increase across various sectors, ordinary practitioners also face the same learning and competitive pressures as top talents.

According to the Maimai report, the intensity of talent competition in the AI field was significantly lower than that in the new economy industry in 2025. However, the supply-demand ratio for AI talents exceeded 1 for the first time, indicating that the AI field has entered a state where the supply of talents exceeds demand.

"Qujie Business" noticed that on current recruitment software, positions in film and television production, marketing, product operations, financial analysis, human resources, and even administrative management all require "proficient use of AI tools," "possess Prompt engineering capabilities," and "ability to optimize workflows based on large models."

"Composite" practitioners with both industry experience and the ability to work with AI are becoming a "new necessity," especially in traditional content industries such as advertising and film and television. When recruiting for its short drama department, Motie Cultural Group requires applicants to be able to "conduct training on prompt words related to short stories and produce content through AI models such as Claude and Doubao." Some companies even establish AI research groups to organize employees in learning to use AI for efficiency improvements.

Embracing AI is the trend, but this learning is also bringing new emotional pressures. A survey by LinkedIn shows that many workplace professionals believe that AI training makes them feel as if they are taking on a second job. AI-related learning and training have not made them feel more capable but have instead increased stress and extended working hours without bringing actual work improvements.

On the one hand, the hallucinations of large models have brought about a significant amount of verification and correction work. Many advertising practitioners have stated that AI can quickly produce visual drafts, but they often contain illogical copywriting, distorted fonts, and human figures. They have to explain to clients why AI deliverables cannot be used directly and repeatedly repair the AI-generated content. A workplace blogger made a vivid analogy that current AI is like a new employee graduated from a top university—diligent, with a good attitude, and able to work overtime, but requiring a significant amount of time for coaching and guidance, and even needing help to fill in gaps and wrap up loose ends. Ten AI tools cannot bring the same business progress as ten people.

On the other hand, "AI employees" are also taking away the jobs of many interns. Data entry and material organization, which used to be tasks for interns, can now be completed by AI in a few minutes. Even programmers, who face higher employment thresholds, are experiencing a sharp decrease in demand for entry-level positions with the widespread use of AI programming tools such as Cursor and Windsurf.

A report by magazine mentioned that in 2025, the number of entry-level positions such as recent graduates, apprentices, and interns in the UK has been declining year-on-year, with coding and engineering positions in the UK decreasing by more than one-third.

The situation is similar in China. According to the Maimai report, from January to October 2025, the recruitment volume for entry-level talents has significantly decreased, with newly released positions requiring less than one year of work experience declining by 39.71% year-on-year, while the demand for positions requiring 3-5 years and over 10 years of work experience remained basically stable. This means that the difficulty for newcomers to enter the industry and for older workers to switch careers is increasing day by day. Industry experience and in-depth knowledge accumulated through long-term commitment may become one of the core barriers for future job seekers.

03. AI spurs new positions

While reducing entry-level positions, the rapid iteration of large models and industrial demands have also spawned many AI-native positions.

According to the Maimai report, 79.55% of Maimai users stated that their companies have already deployed AI, with the degree of AI deployment and application being proportional to company size. The proportion of companies with over 10,000 employees applying AI reached 91%, while the AI coverage rate was the lowest at 66.60% for companies with fewer than 500 employees. Among them, 12.18% of users stated that their companies' deployments have brought significant value.

The earliest positions to emerge with the explosion of large models were those involved in training and feeding AI, such as data annotators and prompt engineers. Data annotation involves manually helping large models understand the world, such as labeling road conditions on rainy and slippery days, distinguishing emotions of characters in images, and even correcting content misidentified by AI. Some large models designed for vertical scenarios, such as medical scenarios, have higher requirements for the professional knowledge and understanding abilities of annotators. Therefore, the recruitment requirements for many high-end annotation positions have been raised to graduates from 211/985 universities, with monthly salaries reaching two to three hundred thousand RMB, regarded by many liberal arts students as the best stepping stone for "transcoding."

However, some low-end annotation positions still suffer from high substitutability, intense workloads, and poor treatment. Some annotators also worry that as annotation accuracy continues to improve and AI becomes smarter, the AI they have trained may eventually no longer need them.

In 2025, besides positions involved in training large models, some emerging positions such as ethical review data governance have also emerged.

The deeper the technology lands, the more prominent the compliance risks and ethical controversies become. Some academics state that the current constraints on AI are significantly weaker than its potential impact, and companies are largely still in a state of "self-regulation."

Recently, OpenAI publicly recruited a "Head of Safety and Risk Preparedness" with an annual salary of 555,000 USD. This position is responsible for assessing and addressing systemic risks that AI may trigger in areas such as mental health and cybersecurity. A report by the Tencent Research Institute shows that such positions often require a background in both technology and law or ethics to ensure human intervention and accountability mechanisms when AI makes mistakes.

Domestically, with tightening policies, companies are responsible for the authenticity and compliance of AI-generated content. "Qujie Business" noticed that many domestic tech companies are currently recruiting for positions related to "AI safety," including two types: one for audit and risk control, and the other for research positions responsible for systemic safety prevention and control, with monthly salaries mostly exceeding 30,000 RMB.

The Tencent Research Institute once pointed out that the new careers spurred by AI are currently unstable, with rapid emergence and contraction following technological iterations. Even AI cannot predict where the next employment trend will be. For job seekers, rather than chasing short-term popular positions, it may be better to accumulate core competencies through in-depth industry involvement and focus on doing their current jobs well. Perhaps in the ever-changing industry iterations, they can find their own stable anchors.