newsJune 4, 2026 2 min read

Amazon's Next-Gen Proteus Takes Plain-Language Orders on the Warehouse Floor

At its Delivering the Future event in London, Amazon unveiled a next-generation Proteus robot that employees direct with conversational language, part of a €10 billion European fulfillment investment. The shift worth watching is the instruction layer — natural language over the hardware — not the robot itself.

Source: About Amazon

A single origin point routing curved paths out to several waypoints across a faint floor grid
CrateOS monitoring note: the news isn't the robot, it's the instruction. When 'go move that' compiles into priority, route, and timing, the routing logic becomes the asset — and the question is who owns it.

On June 4, at its Delivering the Future event in London, Amazon introduced a next-generation version of Proteus, its autonomous mobile robot. The headline change is how workers interact with it: employees direct the new Proteus in plain, conversational language — "no technical commands and no programming interface," per Amazon — and the robot handles the rest. "You tell it what needs to be done. It figures out the priority, the route, the timing," said Scott Dresser, vice president of Amazon Robotics. "It becomes your assistant for material movement." Where the original Proteus is confined to dock areas at 25 U.S. fulfillment centers, moving carts that can weigh close to 400 kilograms, the new generation is designed to work anywhere items move across fulfillment and delivery sites. It is piloting in Amazon's labs now, with European deployment planned for the first half of 2027, alongside a broader €10 billion European fulfillment investment, 25,000 planned regional hires, and companion systems Vulcan (touch-sensitive picking) and STARK (collaborative tote handling).

For anyone running a warehouse today, the durable story is the move up the stack — from teaching robots to telling them. An instruction layer that translates "clear the inbound dock" into priority, route, and timing is an orchestration capability, and orchestration is the part that compounds. The hardware will commoditize: touch-sensitive arms and tote movers will be available from several vendors within a few years. What won't commoditize is the logic that decides which task is urgent, which lane to take, and what "done" means on your floor. The trade-off to watch as fleets like this scale is ownership — when the instruction layer belongs to the robot vendor, your operating decisions quietly migrate into someone else's model. The H1-2027 timeline gives planners room to decide which parts of that decision layer they want to keep tunable inside their own WMS rather than spoken into a black box. The win — fewer people doing strenuous, repetitive moves — is real. Just make sure the dispatcher stays yours.

Your exceptions are costing you.Let's talk.