The reality-check beat: "<5% of sales" meets a retired 30,000-vehicle pioneer; dexterity named the bottleneck
From the Robots digest for 2026-10-11
What happened: A GCN article (2026-10-11, ~4:45 PM ET) reports a Spartanburg, South Carolina car plant retired its first humanoid robot after it helped build 30,000 vehicles in eleven months — with a "more dexterous" replacement taking its place on the line. The capture names neither the automaker nor the robotics vendor. It lands alongside the Oct 10 analysis tier: a The Robot Report op-ed (~8:45 AM ET Oct 10) argues dexterity is physical AI's real bottleneck — that converting demonstrated capabilities into reliably useful work, not demos, is the field's next challenge. A Japanese analysis (translated, ~4:45 PM ET Oct 10) floats a sharper commercial statistic: humanoid "actual work" is reportedly less than 5% of sales in the industry.
What's genuinely new: The Spartanburg item is a development of "robots — Robot aftercare: maintenance, service, and end-of-life infrastructure" — that thread's Oct 6 entry carries Figure's F.02→F.03 retirement ("had built BMWs"; a 75-ton arc furnace at Imatra, Finland) with the standing finding "disposal as marketing, no published decommission standard." What the Spartanburg item adds is what that story lacked: a quantified deployment output (30,000 vehicles across eleven months) and the successor-installation step — deploy, produce, retire, replace, with a performance datum attached, though vendor identity is unconfirmed in capture. It is also the counterweight to the sub-5%-of-sales claim, arriving the same news day: the Japanese piece floats "actual work" under 5% of sales, while a machine ran eleven months on a car line and participated in 30,000 built vehicles. The dexterity op-ed deepens the "Humanoid reality-check war (Agility's humility campaign)" thread, which already carries The Robot Report as a skeptic-side voice (Sept 8 hardware-bottleneck op-ed) — a development, moving from hardware to dexterity. The sub-5% figure stays a single-source, no-methodology claim filed to the thread's price-discovery pattern, unverified. The dexterity argument has sensor-side echo in the same window: Shenzhen LDRobot's tactile-perception partnership and the Japanese food-picking piece both treat soft/fragile manipulation as the open engineering problem.
Watch: Whether the Spartanburg plant's automaker and vendor get named (the aftercare thread's BMW strand is the record's existing retirement-with-successor story); whether the 30,000-vehicle figure gets independent confirmation; what "more dexterous" turns out to mean in spec; whether the dexterity framing picks up benchmarks or named deployments; whether the <5% figure gets a real methodology and a second source; the standing watch items of the reality-check thread (deployment-hours verification, Humility-v-S-1 tension).
Links: GCN — humanoid retired after 30,000 vehicles · The Robot Report — Why dexterity is physical AI's real bottleneck · Note.com — humanoid "actual work" under 5% of sales