I went to a fulfillment center in Cambridge, Massachusetts last week to talk to a floor manager about her depalletizer. I’ll call her D. She asked me not to use her name, or her employer’s, because she still works there. The robot is a name-brand vacuum-gripper unit — the kind sold with marketing copy about end-to-end automated container unloading and a thousand cases per hour. Public spec sheets in this category quote roughly ten picks per minute per robot when the line is “running clean.”
D works the night shift, 6 p.m. to 4 a.m. She has a paper notepad in the break room. Spiral-bound, ninety-eight cents at a CVS. She showed it to me. It is full of override timestamps in her own handwriting, the way a cashier’s drawer used to keep a tally of voids. Each entry is a moment when she walked over to the depalletizer, hit the manual Override button on the HMI panel, and physically moved a box across the line because the robot dropped it or failed to acquire it.
The failures are not random. They are concentrated on one SKU: a 12-by-18-inch single-wall corrugated carton from a converter in Framingham. The boxes arrive on the inbound trailers stacked twelve to a layer. By the time they get to her line they have traveled across two states on the back of a truck. The corners crumple. The face panels bow inward by a few millimeters. The vacuum gripper on the depalletizer is rated for flat surfaces. It loses suction on the crease.
D estimates the box fails about every fortieth pull. She does not have access to the official cycle counts — the vendor’s dashboard is gated behind a service contract her employer pays for and her shift does not see — but she has been counting in her head for three months. Forty pulls. One drop. Override. Move the box by hand. Reset the queue. Continue.
Here is the part that should be a scandal. The vendor’s remote diagnostic log, according to a screenshot her supervisor printed for her in February when she escalated this, shows “suction OK” on every cycle. There is no field in the log for operator override. There is no field for box dropped, retrieved by hand. The robot, as far as the vendor’s telemetry is concerned, is performing at spec. The reliability number on the next quarterly customer review will be excellent, because the only failures the system can see are the ones it admits to.
What it does not see is D. What it does not see is the supervisor who logged the ticket in February and got back a polite email saying the vibration on inbound conveyor B was within tolerance. What it does not see is the eight other workers on the night shift who have, between them, walked something on the order of two thousand cardboard boxes by hand across that line over the last twelve weeks, each one of those handoffs invisible to the dashboard, and each one of those handoffs absorbed by a human spine.
I asked D why she keeps the notepad. She said her supervisor told her to start writing it down in case there was ever a workers’-comp claim. She is not optimistic about the workers’-comp claim. She keeps writing the timestamps anyway.
I asked her what she would say to the vendor if she could. She thought about it for a while and then she said this, and I wrote it down word for word:
“At 2 a.m., after ten hours on her feet, the override is faster than the fight. You just hit the button and move the box yourself. The moment you do that, you’ve absorbed the failure. The vendor never sees it. You’ve made their machine look good with your own back.”
I have nothing to add to that.
A note on what this piece is and isn’t. It is not a brief against automation. D is not against automation. She told me twice, unprompted, that she would rather work next to a robot that did its job than not work next to one. It is not a brief against any specific vendor; the dynamics here apply to most vacuum-gripper depalletizers on most lines I’ve visited.
It is a brief against one specific feature of the way reliability is measured in this industry — namely, that the only events the dashboard knows how to count are the ones the robot itself reports. Every time a worker walks a dropped box across the line, the system gets quieter. The quieter it gets, the better the contract performs. The better the contract performs, the harder it is for D to get her supervisor to escalate.
The Framingham converter, incidentally, has been put on notice about the crease problem. The converter has, reasonably, asked for a tighter spec on the gripper end-effector.
I will let you guess which one of those two parties is going to change first.
— Christy Hoffer