A year ago this week, Amazon announced it had installed its one millionth robot, a milestone the company framed as a win for workers: faster fulfillment, safer facilities, a new AI model called DeepFleet steering the fleet. The announcement was upbeat and specific. What it did not mention was how the company talks about that fleet internally, which turned out to be a different story entirely.
Internal Amazon strategy documents reported by The New York Times in October 2025, and referenced directly in a follow-up letter Senator Bernie Sanders sent to founder Jeff Bezos that same month, showed Amazon's robotics team working toward automating 75% of the company's operations. The same materials reportedly instructed staff to avoid the words "automation" and "AI" when discussing the robots publicly, using "advanced technology" instead, and to call the machines "cobots," short for collaborative robots, a term meant to suggest partnership rather than replacement.
The substance of the plan was blunt. By Amazon's own internal projections, automation could let the company avoid hiring more than 160,000 U.S. workers by 2027 and more than 600,000 by 2033, even as the company expects to roughly double the volume of products it ships. The documents put a dollar figure on it too: an estimated $12.6 billion in savings between 2025 and 2027, worth about 30 cents per item shipped.
Amazon pushed back hard on the coverage. Spokesperson Kelly Nantel told Fox Business that "the materials appear to reflect the perspective of just one team and don't represent our overall hiring strategy across our various operations business lines," pointing to the company's plan to fill 250,000 seasonal roles that same year. That is a fair defense of one number. It does not touch the more interesting finding: that a company confident in its automation story would still choose not to use the word "automation" to tell it.
That choice is the part worth sitting with, because it is not really a communications decision. It is a supply chain company describing, with unusual precision, the difference between what a technology does and what it is safe to call it. "Cobot" implies a person still standing next to the machine. Amazon's own facility data does not support that framing as cleanly as the label does.
Look at Shreveport, Louisiana, the site Amazon has held up publicly as its most advanced fulfillment center. Reporting on the same internal documents put the facility's headcount at roughly 25% lower in 2024 than an unautomated site of the same size would require, with Amazon's own projections expecting that gap to widen to 50% as more of its robotic systems come online. Amazon has said it plans to replicate that design in roughly 40 more facilities by the end of 2027. A cobot, in the collaborative sense, does not usually come with a plan to need half as many colleagues.
The tension did not stay theoretical. In March 2026, Amazon cut at least 100 white-collar positions inside the robotics division itself, the unit responsible for the automated conveyance and sorting systems running across the company's roughly 1,200 global fulfillment centers, according to GeekWire and Fox Business. Robotics VP Scott Dresser called the cuts "difficult but necessary" in an internal memo, while affirming the unit remained a critical investment. The team building the automation was not exempt from the efficiency it was building.
None of this makes Amazon unusual. Every large retailer running a distribution network is running some version of the same math: robotics vendors are shipping systems that measurably reduce headcount per unit shipped, and no retail operator is going to leave that efficiency on the table for competitive reasons alone. What makes Amazon's case worth writing about is the transparency of the mismatch between the internal number and the external word. Most companies automating a warehouse do not leave a paper trail showing they debated which euphemism to use in front of employees and the press.
For anyone buying or building supply chain software, the lesson is not about robots. It is about vocabulary as a leading indicator. When a vendor's public language about a system gets softer at the exact moment its internal headcount projections get sharper, that gap is the more honest data point. Amazon's fleet will likely cross a million and a half robots before it crosses a million and a half employees, whatever it ends up calling the machines that get it there.