Robots don't file complaints

When Amazon announced it had deployed its one millionth warehouse robot in mid-2025, the press release read like a technology milestone. A new AI foundation model called DeepFleet. Efficiency gains of 10%. Robots spanning over 300 facilities worldwide. It sounded like a story about innovation. But leaked internal documents, reported by the New York Times in October 2025, told a different story. Amazon's plan is to automate 75% of its operations and replace more than 600,000 jobs by 2033. That is not a technology roadmap. That is a labor strategy. The question worth asking is not whether robots can do the work. It is why the economics so clearly favor replacing humans, and what that tells us about the actual cost of employing people.

The real ROI calculation

The business case for warehouse robots is usually framed in terms of speed, accuracy, and throughput. But the spreadsheet that matters most is the one tallying up workplace injuries, lawsuits, union negotiations, and employee turnover. Amazon's safety record has been under scrutiny for years. In 2021, the serious injury rate at Amazon warehouses was 6.8 per 100 workers, more than double the 3.3 rate at non-Amazon warehouses. Despite employing roughly a third of all U.S. warehouse workers, Amazon was responsible for nearly half of all injuries in the industry that year. By 2023, the Strategic Organizing Center's analysis of OSHA data found 36,170 serious injuries across Amazon facilities, averaging one serious injury every 14.5 minutes. In December 2024, the U.S. Department of Labor reached a settlement requiring Amazon to implement corporate-wide ergonomic measures across all fulfillment centers, sortation centers, and delivery stations in federal OSHA's jurisdiction. Here is the uncomfortable irony: the robots themselves appear to make things worse for the humans still on the floor. A BBC investigation found that Amazon facilities with robots had injury rates roughly 50% higher than those without. Workers reported being confined to fixed stations, repeating monotonous tasks at machine pace. Picking expectations reportedly jumped from 100 items per hour to 400. The robots did not replace the dangerous work. They intensified it. From Amazon's perspective, this creates a perverse but rational incentive structure. If deploying partial automation increases injury rates and liability, the logical endpoint is not less automation but more, enough to remove the humans from the equation entirely.

Why humanoid

Amazon already has over a million robots in its warehouses. Most of them are low-slung mobile units that ferry shelves around. They work well for specific tasks but cannot climb stairs, open doors, or navigate spaces designed for bipedal creatures. This is the key insight behind the humanoid push. Warehouses are built for humans. Racks, aisles, ladders, pallets, buttons, and scanners are all designed for human height, reach, and dexterity. A humanoid robot can theoretically operate in these environments with minimal retrofitting, what the industry calls a "drop-in" advantage. Traditional industrial robots require custom shelving, specialized automation zones, and significant capital expenditure on infrastructure modification. Humanoid robots skip all of that. They reduce deployment timelines, lower upfront costs, and most importantly, they signal something that a shelf-moving Roomba does not: full replacement is the goal. Amazon has been testing Agility Robotics' Digit humanoid at its research facility near Seattle, and reports from mid-2025 suggest the company is building a "humanoid park" in the U.S. to develop software for humanoid robots that could eventually handle package delivery, with robots designed to "spring out" of delivery vans. The humanoid form factor is not about technical superiority. It is about infrastructure compatibility and the message it sends, to workers, to investors, and to unions.

The union calculus

Amazon's relationship with organized labor is well documented. The company has faced accusations of weaponizing algorithmic surveillance to discourage unionization. A study of the union election at an Amazon warehouse in Bessemer, Alabama, found that the company leveraged the same app workers relied on for overtime offers and pay records to disseminate anti-union messages. Workplace injuries fuel union drives. OSHA citations create regulatory pressure. Turnover is expensive. Every lawsuit, every National Labor Relations Board complaint, every viral video of warehouse conditions adds to the total cost of human labor in a way that does not show up in hourly wage calculations. Robots do not file workers' compensation claims. They do not organize. They do not leak internal documents to journalists. The economics of replacement are not just about productivity, they are about eliminating an entire category of corporate risk. It is worth noting that when the New York Times published its report on Amazon's automation plans in October 2025, the company followed up weeks later by laying off 14,000 middle managers. The message was clear: automation is coming for every layer of the organization, but the warehouse floor is first because that is where the liability is highest.

The digital parallel

The same logic applies in the knowledge economy, just without the hard hats. AI agents, the digital counterparts to physical robots, are following an identical playbook. McKinsey's research estimates that currently demonstrated technologies could automate activities accounting for about 57% of U.S. work hours. Salesforce CEO Marc Benioff has said the total addressable market for digital labor could soon reach the trillions. A CNBC survey found that 89% of senior HR leaders expect AI to reshape jobs in 2026. Digital agents do not need HR departments. They do not require benefits, performance reviews, or severance packages. They do not sue for wrongful termination or file discrimination complaints. Just as humanoid robots address the physical liability of warehouse labor, AI agents address the organizational liability of knowledge work. The framing is always about "augmentation" and "empowering people to do more." And in many cases, that is genuinely what happens. But the underlying economic logic is the same as in the warehouse: if the total cost of human labor includes not just wages but compliance, litigation, and organizational friction, then the ROI calculation for automation gets more favorable every year.

Singapore's version of the same story

Singapore offers a useful lens on this dynamic because the country's labor constraints make the math even more explicit. With a tight domestic labor market and heavy reliance on foreign workers, Singapore has been actively exploring automation as an alternative to importing labor. Singapore businesses are piloting AI-powered robots to address workforce shortages, supported by government programs and robotics-as-a-service models. Reports from late 2025 indicate the government has been weighing policies to phase out foreign workers in favor of automation and robotics. The Singapore warehousing and logistics robotics market is undergoing rapid transformation. But the framing is instructive. In Singapore, the narrative is not about replacing workers, it is about reducing dependency on foreign labor. The political packaging is different, but the economic logic is identical: human labor carries costs beyond wages, including housing, social services, and the political complexity of immigration policy. When a government frames automation as a solution to foreign worker dependency, it makes the labor replacement calculus explicit in a way that corporate press releases carefully avoid.

The social contract question

The standard rebuttal to automation anxiety is that technology creates new jobs to replace the ones it destroys. This has been broadly true across industrial history. The automobile eliminated horse-related jobs but created millions of new ones. But the current wave of automation is different in a way that matters. Previous waves automated specific tasks while creating demand for human oversight, maintenance, and adjacent roles. Humanoid robots and AI agents are designed to automate the human role itself, not just a task within it. When the robot has the same form factor as the worker, and the AI agent can handle the same workflow as the employee, the "new jobs" argument requires more specificity than its advocates typically provide. Amazon itself illustrates the tension. The company claims to have upskilled over 700,000 employees through training programs. Its safety reports trumpet year-over-year improvements. But the leaked documents tell a different story: the long-term plan is to need dramatically fewer humans. There is a genuine case that automation improves safety. Removing humans from dangerous tasks is a net positive. Amazon workers themselves have mixed feelings, not uniformly opposed to robots that take on the most physically punishing work. The question is not whether automation can be beneficial but who captures the value. If robots reduce workplace injuries but also eliminate the jobs, the worker is not better off. They are just no longer Amazon's problem. The injury risk does not disappear, it shifts to whatever less-automated, lower-paying work those displaced workers find next.

What this is really about

The story of Amazon's humanoid robots is not a technology story. It is a story about the true cost of human labor in a system where employment carries legal, regulatory, and organizational overhead that compounds every year. Every OSHA citation, every union election, every wrongful termination lawsuit adds a line item to the cost of human workers. Every algorithmic management controversy, every leaked internal document, every congressional hearing makes the gap between human labor costs and robot labor costs wider. Robots do not file complaints. That is not a feature of the technology. It is the entire point.