The boring robots winning right now: arms, AMRs and stock-counting drones

There's a rule of thumb I offer clients who've just watched a humanoid robot video and want to talk about the future: the more shareable the robot, the less likely it is to be doing real work. It's unfair, slightly. It's also, as of early 2026, statistically defensible. The world's working robot population is measured in millions of units, and virtually every one of them is an arm bolted to a floor, a flat trolley with good manners, or a drone that spends its nights reading barcodes. Let me make the case for the boring machines.

The numbers nobody screenshots

The International Federation of Robotics' annual census puts the global stock of operational industrial robots at around four million, growing by roughly half a million new installations a year — records being set with metronomic regularity. Warehouse robotics adds hundreds of thousands more: Amazon alone operates a mobile-robot fleet in the high hundreds of thousands. Against that, every humanoid pilot on earth totals perhaps a few thousand units doing supervised trials — a category we've examined on its own terms. Both stories are true. Only one of them is at scale.

Arms: the fifty-year overnight success

The industrial arm is robotics' oldest product and its most quietly transformed. Three changes matter. Price: capable six-axis arms and collaborative "cobots" now start in the low tens of thousands of pounds, bringing them within reach of the small workshops that make up most of manufacturing. Vision: cheap cameras plus modern machine-learning models mean arms increasingly locate parts rather than demanding they arrive in jigs, dissolving the old integration costs that used to triple every project. And programming: teaching by demonstration and natural-language interfaces are replacing the week of specialist coding. A machine tool that welds, sands, palletises or tends a CNC machine for three shifts without complaint, paying back in 12–24 months, is not futuristic. It's just good procurement.

The more shareable the robot, the less likely it is to be doing real work.

AMRs: the trolleys that ate the warehouse

Autonomous mobile robots — self-navigating carts and shelf-carriers from firms like Locus Robotics, Geek+ and Amazon's in-house lines — are the biggest robotics success of the decade, and they succeed precisely by refusing to be impressive. No legs, no hands: they carry, and let humans do the picking. That division of labour sidesteps the dexterity problem that still humbles the humanoids, which is why AMR fleets scaled to the hundreds of thousands while humanoid hands were still learning totes. The productivity gains — commonly doubling picker output by eliminating walking — are among the best-documented in automation. Many vendors now sell them as "robots as a service" by the month, part of the broader shift away from big capital purchases that mirrors what's happening in AI pricing.

Drones that count things

My favourite boring category: inventory drones from the likes of Verity and Corvus that fly warehouse aisles at night, scanning barcodes and reconciling stock against the system. Cycle-counting is work humans do slowly, expensively and badly from scissor lifts; a drone does it continuously and hands the discrepancies to a human in the morning. It's a perfect miniature of good automation design — a bounded environment, a repetitive measurable task, and a machine shaped by the job rather than by science fiction.

Why boring wins, and what to do about it

The pattern behind all three categories is the same: constrain the problem until current technology solves it profitably. Generality is expensive; specificity pays this quarter. That will change — the same learning-based control driving the humanoid wave is already making arms and AMRs smarter, and the boring robots will absorb those gains first because they're already deployed at scale, with the workflow and org-chart consequences we've written about separately.

If you run an operation with repetitive physical work, my standing advice as of early 2026: audit your dullest, most measurable tasks; get quotes on the specific machine for that task; insist on a payback calculation under two years; and ignore anything with a face. The robot revolution is real. It just looks like a trolley.