Blog

MODEX 2026 had 1,057 exhibitors. You cannot do diligence on a show floor, but you can ask questions whose answers, and the pause before them, tell you something. Eleven questions, plus a card showing which standard covers what, so you stop citing a performance standard for a safety question.

Acceptance is the hinge. Before it, the vendor must prove the system works; after it, you must prove it does not, and you can lose that right by staying quiet. Six clauses drawn from real filed agreements, plus a 2025 case where a buyer paid $129,725, never received the machine, and recovered only what it had paid.

Two constraints settle this choice before any vendor demo starts: how tall your building is, and how your floor was poured. One architecture leaves 19 feet of your ceiling unused. And the floor standard most contracts cite rules itself out for fixed-path vehicles, in its own text.

Seven published numbers from four robot vendors. Not one states how it was measured, so no two can be compared. Here is what each claim leaves out, and the six questions that make a rate mean something.

Five steps, no special software, and the mistake we made ourselves. Includes the figure showing how one pass mark turned a dataset where nothing succeeds into a 48% headline.

When Tesla recalled 2,031,220 vehicles, the recalled part was software and the fix was a push. From December 2026, EU law holds makers liable for defects caused by updates, and for safety updates they failed to ship. The answer regulators reward is not freezing your model. It is a tested update pipeline you can show an underwriter.

Billions raised, and one published operating number. The flagship warehouse deployment has not disclosed how many robots it runs, seventeen months in. Here is what is actually known.

Kroger wrote off $2.5 billion on automated fulfillment and paid $350 million to leave. The robots worked. The filings show four ordinary causes, warnings visible two years early, and not one published throughput number.

The wage data everyone quotes, done honestly. Payback swings from 9 to 44 months on two assumptions most business cases never write down, and only 3 of 9 scenarios clear the 12-month bar buyers now set.

Two warehouse robot vendors left the market in two years. The clause most buyers rely on is unenforceable in bankruptcy. Here are seven that are worth more, each cited to statute so your counsel can check it.

We built the break-even model and published it. Depending where your quote lands inside the published price ranges, buying pays back in 9 months or never pays back at all. Here is the arithmetic, and what it leaves out.

Each question comes from a problem real buyers reported after deploying robots. Ask them before you sign, get the answers in writing, and watch how the vendor reacts.

Six warehouse jobs, the robot type for each, what it costs, and how proven it is, on one screen of cards. Plus three rules from our own testing to use before you sign.

52 open PushT trials, a success envelope learned from the trials that worked, and a finding: two-thirds of failures were not doing anything visibly wrong. They ran out of time. Every number regenerates from a script.