China Hot Tech @ChinaHotTech
← All posts
Robotics · RaaS · Metrics

AGIBOT Worken: Measuring Robot Labor, Not Just Units Shipped

Worken (有效智能工时 — “effective intelligent work-hours”) is AGIBOT’s (智元) proposed physical-AI value unit: active robot count × autonomous work hours × output efficiency ratio. Co-founder and CTO Peng Zhihui (彭志辉) introduced it on 2026-10-29 at AIGBOT INNOVATION DAY in Shanghai. For overseas fleet buyers and RaaS operators, the point is blunt — units shipped are not productive hours. This is a vendor metric framework, not a downloadable API or industry ISO standard. Vendor share and revenue figures below are flagged with asterisks.

· ~7 min read · China Hot Tech

Bright editorial illustration: AGIBOT Worken scoreboard — formula robots × autonomous hours × efficiency, historical analogy from horsepower and kWh to tokens to Worken, fleet-floor humanoids

Illustration: soft sky / amber / slate — Worken as a productivity meter for robot labor, not a chip die or neon HUD. Formula and positioning from on-stage disclosure; asterisked items are vendor or press claims.

If you buy or operate robot fleets, the spreadsheet that still starts with “how many units did we ship / install?” is the wrong first column. A humanoid that sits in a crate, waits for a teleoperator, or fails mid-cycle is not delivering the same labor as one that runs unsupervised for a shift. Worken is AGIBOT’s attempt to name that gap — a single number meant to sit next to horsepower, kilowatt-hours, and tokens as the physical-AI era’s unit of machine value.

What Worken is

Peng’s on-stage framing borrowed the usual industrial analogies: steam-era horsepower, electric-era kWh, digital-AI tokens — and for embodied / physical AI, Worken. The disclosed formula is multiplicative:

Chinese coverage quotes Peng on the practical distinction: a robot powered on for eight hours has not automatically created eight hours of value, and a high one-shot task success rate does not prove the machine can hold a month of continuous, stable work. Worken is designed to push the conversation from a parameter sheet toward a work-hours sheet.

Can overseas buyers use it today?

As a procurement and RaaS framing tool — yes, in the soft sense. You can ask any vendor (AGIBOT or otherwise) for active-robot counts, autonomous-hour logs, and how they define efficiency. As a product you can buy or call — no. Worken is not a Digi-Key SKU, not an open API, and not a published ISO or IEC standard. The 《2026具身智能产业创新报告》 (Xinhua CEIS + CAICT Shanghai) records it as enterprise exploration, not a third-party industry yardstick.

Worken equation board and fleet comparison: same headcount, different Worken bars; can-use vs cannot-use checklist for overseas RaaS buyers

Same headcount, different Worken — the buyer question the metric is meant to force. Efficiency-ratio measurement remains fuzzy*; no independent audit yet.

Worked examples from Chinese coverage

Press write-ups around the event circulate arithmetic examples that make the three-factor shape concrete (treat the efficiency scalars as vendor/press illustrations*):

Other application scenes cited in Chinese press include Chengdu Kuanzhai Alley visitor service, hotel and retail guide deployments, Chimelong entertainment, ASD retail floor guidance, and Shangri-La hotel front-desk cases. Separately, industrial coverage of Longqi’s Nanchang line has described multi-day G2 “并线” (co-line) runs with high reported task success — useful deployment context, but not the same thing as a audited Worken ledger.

Vendor scale claims at the event*

Alongside the metric, Peng disclosed AGIBOT operating numbers that Chinese financial press repeated (all vendor / Counterpoint-cited, not independently audited here):

Those figures explain why AGIBOT wants a post-shipment metric: once cumulative volume crosses a scale threshold, “how many did we build?” stops answering “how much productive labor did customers actually get?”

What the industry report actually says

The same day, Xinhua’s China Economic Information Service and CAICT Shanghai’s industrial innovation center released 《2026具身智能产业创新报告》. Worken and related deployment cases appear inside it as a lead-enterprise exploration. CAICT Shanghai chief expert Zheng Zhongbin (郑忠斌) is quoted to the effect that such an indicator set moves valuation from hardware sales toward productivity delivery — and gives RaaS / hourly billing a shared measuring stick if the industry ever adopts it.

The report’s own caveat matters more for overseas readers than the marketing analogy: the indicator is still enterprise exploration; whether it becomes a common industry scale requires repeated validation across more scenes. In other words, AGIBOT proposed a ruler; the report did not certify it as the industry ruler.

What changes for a fleet / RaaS buyer

If you evaluate Chinese humanoid or general-purpose robot vendors for factory, metro, hotel, or retail pilots, Worken changes the questions you ask, not the contract you can sign tomorrow:

Keep this distinct from this morning’s ChinaHotTech piece on AGIBOT GO-3 Preview + Lingqu OS — that article covers the deploy-time learning stack and open-source plans. Worken is the commercial measurement layer, not another hardware SKU.

Caveats

Bottom line: Worken is AGIBOT’s bid to make robot labor countable the way tokens made model inference countable. For overseas RaaS and fleet buyers, the useful takeaway is the three-factor question set — active count, autonomous hours, efficiency definition — not a claim that the industry has already agreed on a standard. Keep the asterisks on the share, revenue, and example arithmetic until a third party can audit them.

Sources