As Tencent, Baidu and their peers carry out organizational restructuring for AI, whose job positions will be the first to be reshuffled?

From Tencent to Baidu, AI-driven organizational transformation is evolving from partial trials to comprehensive restructuring.

“A matter of life and death.” On the night of November 11, after Tencent's 28th anniversary internal live stream concluded, Pony Ma's speech quickly spread across social media platforms.

Online rumors claimed he “wore a dark expression the entire time,” and the management had not spoken in such stern terms for a very long period. Remarks about “organizational transformation and personnel adjustment” spread rapidly. He stated that this time “the goal is not to cut costs, but to adapt to the new battlefield of the AI era.” The tide has already swallowed hill after hill, and “among the existing land forces, we must prepare in advance for who can board the ship and whether they are capable of naval combat. Frankly speaking, we are already a bit behind schedule.” Those who choose to adapt to the new playbook “will no longer be coddled, lobbied, or allowed to negotiate terms, or refuse to board the ship until they are promised a management position as in the past.” That same night, the claim that “Tencent is laying off 50% of its staff” went viral on social platforms, and Tencent later clarified that this was a misinterpretation of the speech.

The next day, Robin Li took organizational structure as the keynote of his speech at the Baidu World Conference. He said that AI has not brought the expected economic benefits because enterprises have only replaced the motor without redesigning the factory. Companies that used to organize work by positions and departments will operate more around tasks in the future, breaking down position boundaries and enabling humans and Agents to complete work together. “Positions will not disappear, but they will converge.”

Within two days, the founders of the two internet enterprises both directed their focus to their own organizational structures. “The time has come,” an internet industry insider judged to Digital Intelligence Frontline. In fact, his work in the past two to three years has been accompanied by constant adjustments and fluctuations in enterprises. “We don't wait until everything is ready to act, but move forward while experimenting. There will be successes, and there will also be unintended harms.”

AI has brought impacts to different enterprises. A product manager, with seven or eight AI digital employees, developed two systems in just two weeks, breaking the traditional R&D division of labor; a large internet company split work into tiny tasks for employees to claim across departments, challenging the original management functions; some software enterprises began to reduce middle-management positions and re-evaluate the job value of employees.

The corporate organizational pyramid has started to loosen. Some people have gained the production capacity that used to belong to an entire team, while others find that the value of their original positions is fading. Is this ultimately an AI-driven organizational revolution, or are enterprises leveraging AI to accelerate an adjustment that was bound to happen anyway?

Major Tech Firms Are Taking a Knife to Their Own Organizations

Tencent's current round of organizational adjustment targets the management first. Citing people familiar with the matter, Yicai reported that in the internal speech on November 11, Pony Ma mentioned that some middle-level managers are unwilling to make changes, while front-line employees are full of motivation. He called for a flatter and more responsive middle layer, and encouraged front-line employees to take the initiative to move forward.

But a practical problem arises: if work no longer needs to be assigned layer by layer, what else do middle managers have to do?

At Baidu, an experiment is also underway. This year, when Baidu was preparing for the promotion of a large-scale event, it launched a new workbench built with the company's low-code tools, with almost no polished interface. The workbench split the event into modules, which were further broken down into smaller actions, publicly available for employees to claim, with results submitted and points awarded upon completion. To avoid tasks being too difficult for anyone to take on, every piece of work was split into simple enough parts.

“In essence, we broke down a formerly systematic piece of work into individual actions,” a Baidu employee told Digital Intelligence Frontline. Such actions “can be done by anyone,” not necessarily even by professional public relations staff.

In this employee's view, the first thing this mechanism breaks down is departmental silos. Social media content is light enough to serve as a starting point, bringing people from all departments onto the same platform. The platform, rather than managers, now assigns tasks.

These experiments were not initiated spontaneously by employees. The employee heard that the management has convened multiple meetings this year to push for organizational transformation, and all business departments are required to explore new working methods. “The company must have issued a strict order that we have to deliver something tangible.”

On April 28 this year, Baidu adjusted its job level system, integrating the professional and management tracks below the middle management level into a unified digital job level system ranging from level 5 to level 12. From this employee's observation, one of the key tasks of managers since then has been to explore how organizations operate in the AI era.

Some employees observed that many tasks are not yet fully autonomously scheduled by Agents, but the next step will definitely be for AI to automatically push tasks according to progress.

“You become a fence or handrail for AI. You tell it to go this way, to handle that part, and you teach it how to make this workflow run smoothly.”

Meanwhile, in Baidu's product department, boundaries of other positions have been broken. At the Baidu World Conference on November 12, Chen Yaozong, a DODO product manager, shared that he used to write a little code but dared not submit it directly for launch for fear of causing problems. Now he dares to modify parts of the front end, back end and even the database, because R&D and testing personnel have converted professional specifications into AI review rules, and non-compliant submissions will be rejected.

The DODO core team only has a dozen employees. Chen Yaozong said that if you count all the Agents behind each person, it is like a team of more than 300 people. The internal DODO user group received two to three thousand feedback entries per day at its peak. Every night, AI reads the feedback, checks the code base, simulates the review perspectives of R&D, operations and supervisors, and conducts multiple rounds of questioning on requirements. By four or five o'clock in the morning, the requirement document is already generated. Chen Yaozong checks it at 7 a.m., puts forward revision suggestions on his way to work, and by the time he arrives at the office at 9 a.m., the content to be updated and launched that day is usually already finalized.

At the same time, work that originally belonged to different positions is now being completed by fewer people. The experience of professionals is written into Skills that other employees and Agents can call on.

Employees have mixed attitudes towards this change.

Few people are certain that their positions are secure. Young employees worry about their insufficient seniority; senior employees worry about their high salaries and the fact that their original experience is no longer scarce; middle managers are facing the question of how much of their management value, which used to come from coordination and relaying information up and down the hierarchy, can still be retained. A business leader sighed to Digital Intelligence Frontline that he does not know when he will leave the company, and can only prepare mentally in advance.

“The departments are still there, and the structure is still there. But part of the work has bypassed the original management chain and is flowing on the platform.”

“Everyone has accepted this,” one insider said. Some colleagues are proactive, both in response to the company's requirements and to prove their value in the next round of personnel adjustments. Some take advantage of the company's job and task opportunities to embrace AI, learn to develop products, or try to run social media accounts, accumulating tens of thousands of followers in a few months, leaving an extra path and a sense of confidence for the possibility of leaving large tech firms in the future.

For large tech firms, this is an adjustment that touches on original interests and management relationships. At some small companies, founders have personally stepped in to redesign the entire organizational structure.

The Broodmother and Embryos

One second before answering my call, Wei Jiaxing had just assigned a task thread to AI.

He is the founder of Nanjing Yunfu Intelligence, a company established in 2018 that uses AI voice agents to provide customer service, telesales and customer return visit services for enterprises. Wei Jiaxing has not written code for many years, but he returned to the front line this year. He told Digital Intelligence Frontline that the new project he is responsible for updates about 1,500 feature points a month; the company's R&D team of about 30 people, also using AI programming tools, delivers 500 feature points a month.

“That's equivalent to the R&D efficiency of 90 to 100 people,” and this is only the output of his part-time work.

Wei Jiaxing compares his new role to the Broodmother in StarCraft. “What does the Broodmother do? It produces embryo-level products.”

This year, he personally built 8 sets of internal company systems, covering demand insight, R&D management, marketing growth, CRM, human resources and finance. Each set is a rough initial version, which he calls the “embryo version.” These 8 systems were handed over to 8 AI-savvy executives: the R&D management system to the head of product R&D, the HR system to the head of human resources, the finance system to the CFO, for them to continue iterating.

Since all 8 systems were built by one person, their data has been interconnected from the very beginning. Customer demands enter the system, flow directly to the R&D management system, and are then assigned to engineers' programming Agents; after the code is written, the upgrade log is automatically generated, entered into the marketing system, and published to the company's blog.

“There is no human collaboration in the middle of every pipeline.”

He spent one and a half months replacing all the company's paid SaaS software with these 8 systems. In the past, some parts of the company's commission system were unreasonable, because the off-the-shelf software only supported several fixed configurations, forcing the company to accommodate the system. “I had to design the company's policies to fit the system.” Now he can modify the software according to his own business rules.

Many companies choose to add Agents on top of their original systems. Wei Jiaxing thinks this is “a bit too low-dimensional.” Every data transfer consumes tokens and remains unstable. He hopes to eliminate unnecessary processes directly.

For example, in the past, every version Yunfu completed usually contained about 80 requirements, which required a product manager to read through every requirement document and write an upgrade log, taking about 5 hours. Now AI has access to all the data, and the upgrade log is generated automatically.

“That process has disappeared completely.”

In January this year, Yunfu broke the traditional professional division of labor in the R&D department. The former groups for front end, back end, testing, operations, UI and other functions were adjusted into three full-stack teams, plus a design center and an architecture center.

“If an engineer only uses AI to complete the part they are responsible for, with pre and post processes still remaining, their production capacity cannot be fully unleashed,” Wei Jiaxing said. Letting one or two people with AI take charge of the entire project can boost efficiency “not by 20% to 30%, but by 10 to 20 times.”

He named the new organization a “bubble factory.” ByteDance and Tencent can launch multiple business units and countless Apps at the same time, and eventually find the winning products; now a 100-person company can do the same. Each “bubble” consists of one to several core employees plus a group of Agents, independently advancing a project. Once they deliver the agreed performance, they can get 5% lifetime dividends of the project. “The team gets a sense of internal entrepreneurship, with very little risk.”

However, Wei Jiaxing admits that there are also some problems. After the middle platform was dismantled, code specifications, reuse and precipitation between projects are weakening. What worries him more is the architecture: if the project is not well designed at the beginning, the Agent will keep writing code along the wrong context, and it will become more and more chaotic.

“An engineering team without business thinking, if allowed to independently take charge of a project, could cause disastrous results.” That's why he dare not assign new projects to young people who lack architectural experience, and he writes the “embryo” first before handing it over.

The logic of recruiting new employees has also changed. Yunfu has recruited very few people this year. “It's not that performance is declining, it's that we really don't need that many people.” In the past, people could join the company as long as they passed the written test, starting with simple tasks. “Now there are no simple tasks left for you to do, because AI will finish all the simple work.” He recently recruited two master's graduates from top 985 universities, asking them to partner on a project, one proficient in programming and the other with a good aesthetic sense, both capable of quickly understanding business logic.

The threshold for talents has been raised. What about old employees? Wei Jiaxing admits that so far this year, Yunfu has only laid off two people, and the rest are mostly absorbed by job transfers, even if their original salaries are higher than the new positions. He acknowledges that some people are no longer suitable for their original jobs, but these employees have followed the company for many years and have made no mistakes.

“In the old era, they were competent and professional engineers.”

What confuses Wei Jiaxing is a set of numbers. “If production capacity is amplified dozens of times, but revenue only increases by 20% to 30%, what should we do?” The number of business opportunities cannot increase dozens of times overnight.

The excess production capacity can only be invested in new products, expanding to surrounding areas. But every new product has to start from scratch, which will also distract attention from the main battlefield.

“It's very philosophical,” he said that he has been thinking about how to optimize recently, “maybe in another quarter, I will have a clearer idea.”

Nowadays, this CEO who has become a programmer again assigns 20 tasks to the programming Agent at 11:30 every night before going to sleep. The next morning when he wakes up, he checks the results first.

Who Is Gaining, Who Is Losing

In March this year, the operation and maintenance team of Tuya Smart found an abnormal bill, showing that an employee burned thousands of dollars worth of tokens in one day. The operation and maintenance and HR teams thought someone was using the company's computing power for side jobs, and asked the security department to investigate.

The investigation found that it was a product manager. He ran seven or eight digital avatars at the same time, working for him 24/7, and developed two systems in two weeks. “If we assigned this to the internal R&D team, it might take two to three months,” Liu Longwei, CISO of Tuya Smart, told Digital Intelligence Frontline.

In Liu Longwei's view, the people who use AI best in the company are product managers who know a little about R&D. In the past, product managers who mastered requirements had to fight for manpower and schedule slots from various R&D departments; now, “raising a bunch of AI” is enough.

Tuya also began to restructure along business lines. Departments originally divided by functions such as cloud, APP & front end, and hardware, were transformed into business lines and project teams, led by a small number of core personnel with Agents. Back in March this year, the company also set a rule: no more than two maintenance personnel for each existing application.

In many companies, people who can define requirements are the first to obtain the productivity amplified by AI.

A Baidu employee observed that the company is also recruiting a group of young AI talents from outside, some call them “genius youngsters”: young people who can independently deliver results with a few team members and computing power, and receive good salaries. At the same time, the execution layer is becoming less costly. Almost all enterprises see that the next few years will be a “talent dividend period”, with a large base of fresh graduates, allowing enterprises to recruit excellent students at relatively low costs.

The talent pool is diverging at both ends, while the middle layer is being squeezed.

What is also being redefined is the IT department.

Yang Senlin, Vice President of Zhongshun Jierou, a paper products enterprise, said that the company's IT department no longer accepts any BI report development requests. In the past, IT spent more than 4 million yuan a year on outsourcing and manpower costs to write various reports for business departments; now business personnel use AI to do analysis on their own. “The professional knowledge that used to belong to IT professionals has had its threshold lowered today. So what does IT do? It builds a stable underlying infrastructure and provides technical support,” said Gao Bo, the company's president. The IT department's current work is “not to help generate an analysis or a model”, but to act as a manager of tools, solving technical problems between tools and business operations.

Tuya's R&D team is doing similar work: transforming all existing code and businesses into a unified AI-readable platform. “In the past, it was collaboration between people”, and in the future, interactions between AIs will be built through the knowledge base.

This force will in turn impact software companies. “Some IT departments will definitely have the incentive to write their own systems to replace some paid software, otherwise the IT department will face layoffs,” one insider said.

In addition, the group most impacted by this round of changes will be middle