Amazon plans to replace 600,000 jobs with robots. How will AI reshape the power structure in the workplace?
Editor's Note
According to a recent report in The New York Times, Amazon is accelerating its automation strategy and plans to replace more than 600,000 jobs in the United States with robotic systems in the next few years. Multiple sources familiar with the matter and internal strategic documents show that the company hopes to achieve this goal by 2033. Amazon's robotics R & D team is working on applying automation to 75% of the company's overall operational processes and expects to reduce approximately 160,000 jobs in the United States by 2027.
Amazon's move has once again prompted society to examine and reflect on the structural impact of AI on employment and the workplace ecosystem. Less than three years have passed since OpenAI released ChatGPT, and the new productivity represented by artificial intelligence has profoundly reshaped the operating logic of the workplace. AI is gradually taking over tasks such as repetitive task execution, creative content generation, and predictive analysis, which originally relied on human labor, liberating employees from tedious work.
However, in the face of this technological change, employees' emotions are complex and contradictory. While they sincerely admire the advancement of AI, they are also anxious about how to keep up with the AI wave and worried about the potential job replacement by AI. The anxiety and worry stem from the unknown. What needs to be clarified urgently is what AI truly represents when it is fully integrated into the organization's operation? How can human employees effectively collaborate with AI? And what far - reaching impact will this technology have on the existing organizational structure?
AI Is More Than Just a Tool
In the traditional organizational paradigm, the main focus is on the relationships between "people" and "people," such as the principal - agent relationship and the relationship between managers and employees; and the relationship between "people" and "organizations," such as how employees affect organizational performance through mediators (such as employees' knowledge and skills, employee engagement, etc.).
In the AI era, we cannot simply regard AI as a tool wrapped in advanced technology. The emergence of AI has added a new dimension to organizational relationships - from the two - dimensional "organization - person" to the three - dimensional "organization - person - AI."
The profound impact of this paradigm shift will directly affect the core of management and organizational hierarchy. The existing organizational structure is based on human managers managing other human employees. The core functions of managers are to assign tasks, monitor progress, and evaluate performance. The introduction of AI means that there are non - human "employees" in the organization that can autonomously complete complex tasks. Human managers cannot "manage" an AI agent in the traditional way (for example, through motivational talks or career development planning); human employees also cannot collaborate with AI in the traditional way (for example, by aligning work goals in meetings). At this time, the relationships between "managers" and non - human "employees" (i.e., AI), between human employees and non - human "employees," and even between original managers and human employees will undergo fundamental changes.
Redefining Relationships: Three Modes of Human - AI Collaboration
Professor George Fragiadakis and his team define the relationship between AI and humans in the new paradigm as Human - AI Collaboration (HAIC). The specific relationships can be divided into three categories: Human - Centric, AI - Centric, and Symbiotic.
1. Human - Centric Model:
In the Human - Centric model, humans retain the main decision - making power for tasks. Humans use AI as an enhancement tool to improve their capabilities, and AI cannot replace human roles in relevant task scenarios in the short term. In this model, humans use AI for highly repetitive or data - intensive work.
Typical scenarios: Programmers at companies such as Microsoft and Accenture use GitHub Copilot (a code - assistance software) to write code; radiologists use AI - CAD tools to screen for breast cancer, reducing their workload and improving screening performance.
2. AI - Centric Mode:
This mode refers to using AI as the main agent in the collaboration process, where AI is responsible for decision - making and completes tasks with minimal human intervention (usually humans are responsible for limited problem - solving for escalated issues, result review, etc.). This mode usually features automated operation, and AI can independently execute tasks, but it usually only involves one - way interaction between AI and humans. This mode is suitable for tasks with clear boundaries and where AI capabilities can fully support the work.
Typical scenarios: Waymo's driverless taxis operate fully autonomously in Phoenix, San Francisco, and Los Angeles in the United States, accumulating a total of 96 million miles (without safety drivers) as of June 2025; Walmart collaborates with service provider Symbotic, and an AI scheduling system controls hundreds of mobile robots for autonomous warehousing, depalletizing, picking, and palletizing (humans only appear in abnormal situation handling and upstream quality inspection).
3. Symbiotic Mode:
The Symbiotic mode is a balanced partnership where humans and AI enhance each other's capabilities. This mode is characterized by two - way interaction, joint decision - making, and continuous feedback exchange. This mode is particularly suitable for complex tasks and will be the main cooperation mode between humans and AI in future work scenarios.
Typical scenarios: OMD 180 (a creative service agency) collaborates with Adobe Firefly to repeatedly complete the "prompt - review, finalize" process of human - machine interaction, creating 540,000 brand images in 12 artistic styles in 5 days, achieving human - machine co - creation; SAP customers use AI Copilot Joule to enable employees and AI to collaborate in core processes such as supply chain, procurement, and finance. Employees make business context judgments, and AI performs calculations and process generation.
HAIC (Human - AI Collaboration) is an inevitable trend in the workplace. Its introduction brings new changes to the traditional organizational paradigm and provides new development directions for organizational strategic process reshaping, structural setting innovation, talent management, and motivation.
Strategic Process Reshaping: From "Replacing the Motor" to "Reconstructing the Production Line"
When an organization introduces AI into its business processes, if it adopts the Human - Centric Model and AI - Centric Mode cooperation paradigms, the impact on the strategic process is relatively small. In the Human - Centric Model, human decision - making still plays a central role. The scope of responsibilities of human positions/roles may expand, and the number of process steps may decrease, but it will not bring about a holistic and complex process reshaping.
Organizations that adopt the AI - Centric Mode are often in new AI - native scenarios or new attempts to fully AI - enable existing business scenarios. Such scenarios generally have clear business boundaries and do not pose significant challenges to the overall process of the original organization.
However, if an organization hopes to introduce AI through the Symbiotic Mode, a complete process reshaping is required. Single - point or isolated AI transformation often cannot meet expectations. A research report from BCG and MIT Sloan School of Management points out that if AI is simply inserted into the work scenario without process transformation, the benefits it can bring are only one - fifth of those achieved through process transformation. Comprehensively transforming the existing processes to embrace the new productivity is an inevitable stage. Historically, the pursuit of advanced productivity has often been accompanied by comprehensive process reforms.
Economic historian Warren D. Devine, Jr. described the slow evolution of the replacement of steam power by electricity in "From Shafts to Wires: Historical Perspective on Electrification." In the late 19th century, the textile industry was a pillar industry of the Industrial Revolution and one of the largest users of steam power.
In a typical textile factory, the power core was a huge steam engine located in the center of the factory. This steam engine drove the main drive shaft, transmission belts, and pulleys on the ceiling through a complex system, and then distributed power to hundreds of looms. In the early 1890s, DC motors began to appear in the manufacturing industry. As a new form of productivity, they were favored by business owners for their cleanliness, stable power, and ease of control. The first wave of electrification transformation began at that time. In this round of transformation, most of the transformation plans involved replacing the steam engine with a huge motor in the original location and then connecting it to the existing, unchanged belt and drive shaft system to drive the factory's looms.
This stage of reform was hardly a complete success. The simple replacement of the steam engine by the motor did not directly lead to an increase in productivity, and the production cost did not decrease due to the adoption of the motor. Instead, due to issues such as the adaptation of the overall machinery and limitations in large - scale power supply, the production cost increased. It was not until the 1920s that a new generation of engineers realized that the real electrification revolution was not just about replacing the motor. It was necessary to completely abandon the huge central drive shaft, supply power independently to each machine, and redesign the entire factory's production process based on this to truly break away from the previous steam - engine work paradigm and leverage the advantages of electricity.
Just as it was on the eve of the replacement of steam power by electricity, what about the present night when AI is on the rise?
Lareina Yee, a partner at McKinsey, and her colleagues believe that although AI seems powerful, it cannot directly improve the overall organizational workflow and achieve AI - enabled business transformation without human assistance. If an AI - enabled transformation project focuses on the entire work chain (covering various aspects such as personnel, processes, and technology) from the start, it will have a greater chance of success. Understanding the role that AI can play in each link is the key path to realizing the value of AI - enabled transformation. Redesigning the workflow is an important starting point, which helps to find the key entry points for systematically solving problems, enabling efficient collaboration between AI and humans, and thus more effectively achieving business goals.
Future Organizational Blueprint: Centralized, Flat, and Task - Centered
1. Organizational Centralization
In the process of an enterprise's pursuit of overall AI - enablement, there needs to be a role (such as an AI CoE, i.e., an AI governance team/AI Center of Excellence) that can control centralized AI governance and decentralized execution. The AI - enabled transformation of an organization is different from general innovation practices. Innovation - related work often generates creativity through a bottom - up approach. However, if an enterprise's AI - enabled transformation is carried out in a distributed, bottom - up manner, it may limit the final effect of the transformation and cause repeated investment of resources. Without a centralized AI transformation driver, an organization is likely to fall into the following three traps:
Trap 1: Looking for nails with a hammer. When the AI wave hits, each department has a strong demand for AI - enablement. To pursue the efficiency of rapid deployment, they often choose existing standardized solutions in the market as their first option. At this time, AI helps the enterprise solve the problems that "standardized solutions can solve" rather than "the problems that the enterprise needs to solve." Therefore, it is difficult for an organization implementing AI - enabled reform in a distributed manner to obtain an overall optimal solution.
Trap 2: Duplication. Different departments may sometimes solve similar problems, but departmental barriers often lead to redundant construction. Take the demand for AI - powered customer service as an example. In a large organization, the human resources department needs to use AI to answer employees' questions about many policies, and the finance department also has similar needs. The AI capabilities required for both are similar, but the human resources department and the finance department often design independent solutions instead of sharing AI capabilities collaboratively.
Trap 3: Contradiction. When different departments use different data sets and different models to solve similar problems, they often get contradictory answers. Graham Kenny and Kim Oosthulzen gave a vivid example in Harvard Business Review. In the process of implementing AI - enablement, a multinational bank headquartered in Australia used a risk - management AI in the finance department that marked a specific customer group as high - risk based on traditional credit scores and historical loan performance. At the same time, the customer - acquisition AI in the marketing department identified the same customer group as the main target customers based on digital behavior and social media data. This contradictory conclusion caused serious internal consumption.
2. Hierarchical Flattening
Hierarchical flattening is a typical feature of an organization in the new AI paradigm. On the one hand, AI significantly enhances the capabilities of human employees and expands their scope of responsibilities. With the overall scope of responsibilities remaining unchanged, the number of employees will decrease. The overall reduction in organizational scale leads to organizational flattening. On the other hand, the boundaries between ordinary employees and middle - level managers, and between middle - level managers and senior managers are blurring. Edwige Sacco, the head of the labor innovation department at KPMG, once stated that ordinary employees are liberated from simple work with the help of AI and have more time to focus on more difficult tasks (such as preparing for meetings or strategic dialogues); middle - level managers are freed from tasks such as methodology preparation and spend more time dealing with customer problems and strengthening customer communication. These changes make them closer to the roles of the next level, and the traditional hierarchical barriers gradually disappear in the process.
The core work of traditional managers is to assign tasks, monitor progress, and evaluate performance. In an organization in the AI paradigm, the core work of future managers will be to form mixed teams composed of humans and AI, set strategic goals and ethical boundaries for the teams, and integrate the final outputs. This requires a new set of skills: less direct supervision and more systematic thinking, strategic scenario setting, and abnormal situation handling capabilities. This transformation is likely to give rise to a flatter organizational structure, with a reduced number of traditional managers, perhaps replaced by a new class of "human - AI collaboration orchestrators."
3. Task - Oriented Organization
A task - oriented organization is an organizational form that focuses on the end - to - end solution of tasks and sets organizational roles (human employees & AI employees) according to the task - solving loop. Task - oriented organizations are generally small in scale and flexible in setting, and they have advantages in decision - making, information flow, and resource allocation. Asha Sharma, the Corporate Vice President of the AI Platform at Microsoft, and Yang Guongan, the Dean of Tencent Ivy League, both believe that task - oriented organizations will play an important role in the AI era.
Asha Sharma believes that the traditional organizational structure is a hierarchical system based on "people," while in the AI era, the core of organizational construction is "tasks." The traditional organizational chart (Org Chart) will be replaced by the work chart (Work Chart)/task network (Task network). Organizations will be arranged according to measurable task loops (Task Loop) rather than fixed functions (Lane). Yang Guongan also holds a similar view. He believes that task - oriented organizations are based on future opportunities and emphasize the Owner (rather than the Leader). The traditional hierarchical organization has clear levels, clear and standardized division of labor, and is suitable for a stable environment, while task - oriented organizations are suitable for an environment full of uncertainties.
The change in the organizational structure in the AI era will not happen overnight. It may start in the form of pilot projects/special zones in the organization and gradually complete the innovation. Organizational centralization, hierarchical flattening, and the creation of task - oriented organizations are not brand - new organizational topics. However, the emergence of AI has provided the technological basis for realizing these organizational directions and also provided an opportunity for reform to break the original organizational rigidity.
Repricing Your Job
In the general direction of task - oriented organizations, whether it is the Human - Centric Model, AI - Centric Mode, or Symbiotic Mode, there will be a greater emphasis on measuring task results and designing short - term compensation incentive plans based on task results.
Within the current technological boundaries, the tasks completed by HAIC (Human - AI Collaboration) generally have SOP (Standard Operating Procedures), definite results, and clear result - measurement standards (evals, which are the evaluation methods and standards that need to be carefully prepared before the launch of an AI product/function). Therefore, there is a basis for implementing incentives directly based on task results. At the same time, in the Symbiotic Mode, the measurement standards for human employees will also shift from the relatively broad OKR/KPI indicators to the task - level SLA (Service Level Agreement), such as latency, throughput, and product availability.
There is a comparable case for the greater emphasis on measuring task results in the AI era. Bret