Amazon Has More Than a Million Robots. What Happens to Human Workers?

Introduction

Amazon is turning its fulfillment centers into complex systems operated by mobile robots, robotic arms, software, and people. The company says it already has more than one million robots working in its warehouses, while its 2025 annual report lists more than 1.5 million full-time and part-time human employees. That combination is a useful corrective to the simple idea that warehouse automation means robots replacing everyone. For now, it means that human work is being reorganized around machines.

Key points

• Mobile robots remain the backbone. After acquiring Kiva Systems in 2012, Amazon accelerated its use of robots that bring storage pods to employees rather than forcing workers to walk through aisles to find products. Hercules and Titan are examples of these systems. Amazon’s Sequoia system coordinates robot swarms and is designed to make identifying and storing items faster.
• Robotic arms are moving into more complicated tasks. Systems such as Robin can analyze a cluttered pile of packages, choose an item, pick it up with suction cups, and place it for further handling. Cardinal uses suction to lift boxes and stack them in rolling carts. Amazon expects to more than double its deployed robotic-arm fleet this year, although the current number is still in the single-digit thousands, far below the number of mobile robots.
• AI is becoming a traffic manager. Large groups of robots moving through narrow warehouse lanes can create bottlenecks. Amazon’s DeepFleet suite of AI foundation models is intended to predict congestion and improve routing and coordination.
• Humans still provide flexibility. Warehouses contain products with different shapes, positions, packaging, and order combinations. Robots remain less capable than people at handling every unexpected situation, so employees continue to perform work that requires judgment and adaptation.

What changes for people?

The strongest argument for automation is that it can take on work that is “dull, dirty and dangerous.” The US warehousing sector has generally recorded injury and illness rates above the average for all industries. Robots that move heavy loads or bring storage racks to workstations could reduce walking, lifting, and other forms of physical strain.

That benefit is not automatic, however. Earlier forms of automation often removed a hazardous task from the worker’s immediate surroundings. Collaborative robots change the relationship because they operate near people and can influence the speed of the workflow. Brian Greenwood, a professor at George Mason University, told Ars that this physical proximity and pace-setting ability create a different set of risks. A system intended to make work easier can also create pressure if employees are expected to match the speed of robotic equipment and conveyors.

Amazon’s robotic-arm development illustrates why software matters as much as hardware. Early versions of Robin relied on hand-designed picking rules. A rule that favored large packages could fail when smaller packages covered them, while a rule that favored items near the top of a pile could miss partially buried or otherwise unreachable objects. Amazon’s researchers later trained a machine-learning model on more than 394,000 picking actions from deployed arms. In one reported trial, the best machine-learning method achieved a 93.73 percent success rate, compared with 92.28 percent for the strongest traditional algorithm. A small percentage improvement can matter when a system performs tasks at very high volume.

Why it matters

Amazon’s example shows that warehouse automation is not only a race to acquire more machines. It is also a race to build systems that can perceive surroundings, coordinate fleets, learn from failures, and work safely near people. Human roles may therefore shift rather than simply disappear. Employees could spend less time on repetitive walking and lifting, and more time handling exceptions, monitoring equipment, maintaining processes, and completing tasks that robots cannot yet perform reliably.

Whether that transition improves working conditions will depend on how companies set productivity targets, provide training, and manage safety. The number of robots is not, by itself, a measure of progress. The more important questions are whether workers experience fewer injuries, whether the pace remains reasonable, and whether they receive the skills needed for the jobs created by automation.

Source: Ars Technica AI