Tomas David Ye

Tomas David YeSep 17, 2026

Robot utilization: 91%. Pick rate: +18%. Orders shipped: basically unchanged. Everyone initially thought the problem was downstream. It wasn't. The robot was doing exactly what we'd told it to do. You can buy a robot. You can't buy the logic that makes it useful. The warehouse had moved to goods-to-person. AMRs were bringing inventory to the workstations instead of sending people walking through the aisles. The robots were fast. But then we watched the operation run and something felt off. The bin-placement logic had been built around the old human workflow. A picker could grab three nearby SKUs on one walk through a zone. The robot was making three trips because the system wasn't thinking about robot travel the same way it thought about human movement. Then the cartons reached packing in an order that made sense to the algorithm and not much sense to the person packing them. So the dashboard looked healthy. Robots were moving. Pick rates were up. People were still waiting, rearranging cartons, sending robots back out. Nothing was technically broken. The automation was just executing yesterday's logic faster. That's the part of goods-to-person that gets missed. The robot handles movement. The software decides what moves, when it moves, and in what sequence. If that logic was designed for a warehouse before the robots arrived, you're not automating the operation. You're automating the old one.

Tomas David YeSep 17, 2026

We saw a 14% override rate on our routing system and treated it like a failure rate to eliminate. So we made overrides harder. Added approval steps. Made people justify each one before the system would let them deviate. Use the tool. Stop doing your own thing. Throughput dropped 8%. A picker finally explained it to me directly. "When I override, it's because you don't see what I see. The scanner at station 4's been broken since Tuesday. That aisle's being restocked right now. The bin you're sending me to is actually empty, and has been all morning." The system was mathematically right, every single time. It was also operationally blind, every single time, to things a person standing on the floor knew within the first five seconds of looking at it. Every override we'd been trying to eliminate was someone quietly compensating for something the system had no way of knowing. We weren't stopping bad decisions. We were slowing down good ones, and calling it discipline. We flipped the approach. Made overrides faster instead of harder, one tap instead of a justification form. Throughput jumped 12%. Not because people suddenly started making better calls. They'd been making good calls the whole time. We just stopped forcing them to fight our own system to act on something they could see and we couldn't. If your team keeps overriding the system, the system isn't finished. It's just confident, and confidence isn't the same thing as complete. The person hitting override was never the problem. They were the only one in the building who could actually see the thing your dashboard couldn't.

Tomas David YeSep 16, 2026

At what point does "operator judgment" stop being a person and start being a business rule you never bothered to document? An operator overrode our carton recommendation 500 times. Same SKU. Same fix. Six months. We spent three months analyzing the data to find the pattern. It took four minutes to ask her why. There was a packaging quirk the algorithm never knew about. She'd known it since week two. She'd been correcting it silently—500 times—while we watched the data and called it operator error. Around override fifty, we stopped logging what she was doing. We started asking what she already knew. Turns out she wasn't failing the system. She'd already figured out it was incomplete. She'd been teaching us the rule we forgot to ask for, one shipment at a time. That decision? Saves or costs cents per order. Scale it to fifty thousand shipments weekly and suddenly "boring" is worth more than the person making it. Every repeated manual correction is a message: there's a pattern you missed. Stop retraining. Start asking. The operator isn't correcting the system. She's showing you the rule you never bothered to write down. Are you listening?