ByteDance officially fires the first shot of management reform

Just last night, Liang Rubo sent an all - staff letter to ByteDance's global employees, updating the company's ten leadership principles for the first time in four years.

When the news came out, what the outside world discussed the most were expressions like managers going to the front - line and assessing actual output. But in my opinion, the most thought - provoking principle is the re - emphasized "Context over Control".

This phrase is not new to those familiar with ByteDance. Since Zhang Yiming's era, it has been a signature expression in ByteDance's management concept, which has been around for almost a decade. In the past, it was more like a cultural advocacy, an idea mentioned in new - employee onboarding training. No one would be directly affected for not achieving it.

This time is different. The new version of the principles clearly states that it will be incorporated into the core criteria for managers' promotions and annual assessments. Those who fail to meet it will have their job responsibilities and authorities adjusted. The change from soft advocacy to hard constraint is not because managers suddenly understood the truth. Instead, both the organization and the entire industry have reached a point where they have to face the problems directly.

01 From cultural advocacy to a hard assessment indicator, the old principle encounters new problems

The origin of "Context over Control" can be traced back to Netflix's culture handbook. Zhang Yiming officially shared it as ByteDance's management concept at the Ma Club meeting of Source Code Capital many years ago.

The core logic is simple: as the company grows, instead of relying on layer - by - layer approvals and complicated rules to control risks, it's better to synchronize enough context information to the team and let those who are close to the action make decisions.

ByteDance's early rapid expansion benefited greatly from this approach. The open and transparent OKR system, flat reporting lines, and minimal process approvals allowed the team to move fast during the mobile - internet dividend period. At that time, the whole company was moving forward, with rapid business growth. Problems were masked by the growth rate. Even if there were some setbacks in management control, no one really cared.

Changes occurred gradually. As the business scale expanded, the number of team members increased several times, and diversified businesses spread globally. To reduce risks and align goals, various processes and control nodes naturally increased. Many teams gradually developed complicated reporting lines, multi - level approvals, endless synchronization meetings, and countless reports.

The daily work of managers gradually changed from solving specific problems to transmitting information, aligning stances, and controlling processes.

The "everyone is busy but with no actual output" directly pointed out in the all - staff letter is a common situation in many large companies.

In May this year, the "2026 State of the Team Report" released by Atlassian surveyed more than 12,000 knowledge workers and 173 executives from Fortune 1000 companies globally. The final figures were astonishing: only Fortune 500 companies pay $161 billion annually for "work on top of work", such as status synchronization, repeated communication, cross - departmental coordination, and finding the latest version of documents.

The vast majority of respondents said that everyone on the team is constantly in execution mode but can't find time to do really important things.

ByteDance is of course not immune to this rule. Listing formalistic work as a negative management case this time shows that the internal idle - running problem has accumulated to a level that requires intervention.

Choosing to elevate "Context over Control" to the level of an assessment standard at this time also has a more direct background: the entire industry is transitioning from the recommendation era to the AI era, and the new competition rhythm doesn't allow for a slow decision - making chain.

In the past two years, ByteDance has made significant investments in areas such as large models, multi - modality, and embodied intelligence. The adjustment of the Seed team and the integration of business lines are essentially preparations for the competition in the AI era.

The iteration speed of AI business and the pace of technological change are much faster than those of traditional internet businesses. If a decision has to go through three or five levels of approval to be implemented, the opportunity window may have already passed. At this time, reducing control and delegating decision - making power forward is no longer just a cultural concept but a necessity for business competition.

03 Addiction to control is a common problem across the industry

ByteDance's adjustment this time is not at all out of place in the technology industry this year. In the past six months, from Silicon Valley to China, almost all leading technology companies have been doing the same thing: compressing management levels, squeezing out management bubbles, and bringing decision - making closer to the front - line.

The most typical example is Meta. In May this year, despite setting new records in both first - quarter revenue and net profit, Meta still launched a new round of organizational restructuring, laying off about 8,000 positions and transferring 7,000 employees to the newly established AI department.

Zuckerberg's idea is clear: the company needs to fully shift to AI, and budgets and personnel quotas should be tilted towards computing power and technology. Traditional middle - management and coordination positions should be compressed as much as possible. The company is implementing a flatter small - team model to improve decision - making efficiency.

Microsoft's actions are even more radical. According to Business Insider, Nadella quietly abolished the senior leadership team that had been in operation for decades this year. This power center consisted of more than a dozen executives who reported directly to the CEO. It was replaced by a smaller decision - making group. He personally reviews AI indicators every week and holds meetings with the computing - power team every two weeks.

In the past, Microsoft's business lines operated independently, with their own budgets and KPIs, resulting in high cross - departmental collaboration costs. Now, to fight an overall AI battle, the first thing to break down is the wall of management.

Large domestic companies are also busy. Baidu's Mobile Ecosystem Business Group completed a new round of organizational restructuring in early June, merging business units and upgrading independent businesses. The core goal is to flatten the hierarchy and enable faster integration of AI capabilities and business scenarios. Alibaba has made continuous adjustments in the past three months, from establishing the ATH Business Group at the beginning of the year to setting up a technology committee and splitting new business departments. It has also been repeatedly sorting out the problems of resource dispersion and low - efficiency collaboration.

Recent reports mention that in the first quarter of 2026, the number of layoffs in the global technology industry exceeded 80,000, almost reaching the highest single - quarter record in two years. About half of the job cuts are directly related to AI and automation. Not all of the laid - off employees are front - line workers; a large part are traditional middle - management positions, those responsible for information transmission, process execution, and cross - departmental coordination.

The example of Cloudflare is more representative. The company laid off 20% of its employees this year and offered very generous severance packages. The official statement clearly said that this was not a routine performance optimization but a structural adjustment. In the past three months, the internal AI usage in the company has increased by 600%. A large number of AI Agents have started to undertake routine business processes and cross - departmental collaboration work. The value of the original "middleman" positions is being rapidly diluted by technology.

This is actually the root cause of the management alienation in the industry.

When the business is growing rapidly, companies are willing to pay for management. A few more layers of structure and more processes are acceptable as long as they can stabilize the overall situation, and the costs can be covered by growth. But when the growth rate slows down and AI technology starts to take over standardized process work, those management links that don't directly generate business value become the first bubbles to be squeezed out.

Many teams end up in a strange circle: managers are busy building more perfect systems, more complicated processes, and more intensive reporting systems. Everyone seems busy, and their daily schedules are full. But no one can clearly say what value these works bring to users and business.

Management has gradually changed from a tool to serve the business to an end in itself. Everyone is working for the process and has forgotten the original purpose.

03 It's easy to be serious, but it's always difficult to truly implement

Many companies talk about "Context over Control", but very few can really implement it consistently. Because this is essentially against human nature, especially against the intuition of managers.

For most managers, control power is a source of security. Approval power, reporting lines, and rule - making power can clearly define the boundaries of management and let managers clearly perceive their own value.

Implementing "Context" means delegating decision - making power to the front - line, giving up some control power, and taking responsibility for the front - line's decision results. This feeling of letting go is far less reassuring than personally approving and modifying plans.

This is why many companies constantly talk about delegating power but end up going back to the old way of layer - by - layer approvals. They say they trust the team, but when something really happens, they can't help but intervene, inquire, and make decisions. Otherwise, they feel that their management is not in place.

Over time, the front - line team also learns to push everything up. Anyway, the leaders will take the responsibility if something goes wrong. In the end, the responsibility returns to the managers, and delegating power becomes an empty promise.

This time, ByteDance also lists "going deep into the front - line" as an independent leadership principle, which is actually to supplement the premise of "Context". Managers need to have enough real - world context before they can delegate power. If they sit in the office every day looking at reports and listening to briefings, the information they get is processed through multiple layers. They can neither judge the right or wrong of front - line decisions nor dare to really delegate power.

However, going deep into the front - line is also easy to go astray.

If it's just a simple requirement for managers to go to the front - line a certain number of times a month and write a certain number of research reports, it will quickly turn into a new form of formalism and become another KPI to deal with. Real immersion means going to the front - line with problems to find answers, directly facing the real pain points of users and business, rather than a routine inspection.

This scale is difficult to quantify with a system. In the end, it depends on the managers' own awareness.

There is also an unavoidable problem: the balance between long - term value and short - term assessment. The newly added principles of "doing high - level things" and "daring to set high goals" require managers to look beyond short - term monthly data and focus on long - term high - value tracks.

The principle is correct, but in terms of assessment, how can long - term results be evaluated? How can we distinguish whether a manager is planning for the long - term or just slacking off in the name of the long - term? If the final assessment criteria are still short - term revenue and data, people will naturally return to the habit of chasing monthly indicators.

Looking back, this update of the ten principles is not just a change to a single principle but a set of mutually supporting combinations. To reduce control, managers need to go deep into the front - line to obtain real information; to dare to delegate power, managers need to focus on long - term goals and not get entangled in daily trivialities; finally, all requirements should be reflected in actual output, and business results should be used to verify right or wrong.

This logic is sound, but the implementation is not easy at all. The organizational inertia of large companies is very strong. A management model that has been in operation for many years will not be completely changed by just sending an all - staff letter. There will be setbacks, deformations, and many ways to cope with policies. But at least, putting the problems on the table and clarifying the standards is the beginning of change.

Looking back, the management evolution in the technology industry over the decades has always been about finding a balance between control and flexibility.

The hierarchical management inherited from the industrial era has helped many companies achieve large - scale expansion, but it has also brought side effects of rigidity and bureaucracy. In the internet era, many companies tried to solve this problem with flat and transparent management. But once the scale grows, the inertia of control will pull the organization back to the old way.

The emergence of AI technology now adds a new variable to this old problem. As more and more process - based, standardized, and coordinated work can be replaced by technology, the value boundaries of traditional middle - management are being redefined. If managers still stay at the stage of information transmission and process control, it's only a matter of time before they are eliminated.

ByteDance's serious implementation of the old principle this time is more like an industry signal: the value of management should ultimately return to business output. All processes, systems, hierarchies, and controls should ultimately serve to create user value, rather than becoming an end in themselves.

Of course, no management principle is universal. How far ByteDance's adjustment can go in the end depends on subsequent implementation and iteration. But for the entire industry, it's always a good thing when more and more companies start to reflect on the cost of management itself and squeeze out the water in management.

This article is from the WeChat official account “New Vision” (ID: xinmouls). The author is Li Xiaodong, and it is published by 36Kr with authorization.