AI Agents: Cool Demos vs Real Revenue — What Western E-Commerce Can Learn from China

Watch this exclusive presentation recorded live at the E-commerce Berlin Expo 2026! Gain valuable insights and explore key discussions from industry experts that will help you stay ahead in the world of e-commerce.

Speaker: Björn Ognibeni, Practical Visionary, Co-founder, ChinaBriefs.io & XRLab@MCM

Full title: AI Agents: Cool Demos vs Real Revenue — What Western E-Commerce Can Learn from China’s No-Bullshit Approach

Presentation description: While Silicon Valley perfects Agentic AI demos, Chinese platforms are already deploying AI at scale — and how to apply China’s result-first mindset to your own e-commerce strategy.

(From the E-commerce Berlin Expo YouTube channel description.)

A 28-minute keynote recorded live at the E-commerce Berlin Expo 2026 (Berlin, ~late February 2026; published to YouTube 11 May 2026), delivered by Björn Ognibeni — Hamburg-based “practical visionary,” co-founder of ChinaBriefs.io and the XRLab@MCM at the University of Münster, and lecturer at UC Davis (“Rethinking Digital” course). Ognibeni’s premise: the only worldwide ecosystem not influenced by Silicon Valley is China — the political firewall makes it the natural petri dish for different ideas, which is why Western firms repeatedly get blindsided (DeepSeek, electric cars, Shein/Temu, and now agentic commerce). Auto-generated English transcript, ASR-cleaned. The talk closes with a Q&A (~21:00–28:22, two audience questions).

TL;DR

A practitioner’s no-bullshit reframing of Western enterprise AI through the China lens. Ognibeni argues that the West chases AGI (one super-intelligence that will justify the spend) while China pursues AI diffusion (industrial-scale deployment that already generates revenue today). The talk supplies a half-dozen live cases that the wiki has not previously held in concrete form:

1. ByteDance is the largest AI company worldwide by usage — “50 trillion tokens every day, more than Google AI, OpenAI, and Anthropic combined.” And it’s not categorisable: it’s a social network (TikTok) + e-commerce ($670B GMV vs Amazon’s ~$830B) + AI technology + productivity suite all at once. “It looks pretty totally alien to us. And that’s the reason why China is so interesting — it’s a kind of time machine where we can look into our own digital future.” This is the strongest single anchor the wiki holds for the firm-as-platform category and a direct empirical update of enterprise-ai-adoption’s omniscaler picture.

2. Xiaomi’s dark factory — AI-first manufacturing as concrete instance. “1,123 days from announcement to launch of the SU7 electric car. The factory is so dark because there are no people in there. Stuff goes in, cars come out, 40 cars per hour autonomous production.” Ognibeni’s blunt framing: “That’s AI first. It’s not about the [bullshit] we often talk about here.” Compresses strategic-foresight’s strategic-agility microfoundation into a visceral image — and crucially, locates AI-first not in the chatbot but in the production system.

3. JD.com Joy Streamer — agentic AI generating real revenue today. Virtual digital-twin live-stream hosts for JD’s e-commerce platform. “2.3 billion RMB — about $250M — in sales during the last Double 11 [11 November] season. They are more successful than 80% of their human hosts.” This is not a pilot, not a sandbox — it’s a deployed agentic system at material revenue scale. First wiki source naming a specific RMB revenue line attributable to an AI-agent product.

4. Alibaba’s AQ — AI agent that builds physical supply chains end-to-end. A buyer prompts AQ (“I want to sell Christmas slippers”); the agent does market research, generates design suggestions, matches the buyer to manufacturers who can produce them, and runs cross-language chat (German↔Mandarin) where “Alibaba guarantees the translation works — if there’s a mistake, it’s Alibaba’s fault. That makes you actually use it for live business transactions.” The trust-guarantee on the translation step is what turns a demo into a deployable channel.

5. Alibaba’s Qen one-sentence-purchase agent — agentic commerce at user scale. “Book me a cool hotel in Hangzhou and the train to get there, and buy me an umbrella or a hat depending on the weather.” Today, served via Alibaba’s ecosystem (Fliggy travel + map + weather + e-commerce + payment) with tokenized mandates as guard rails. “This Qen agent can be used by more than 100 million people today — it’s not a test, it’s something live and getting actually used.” The wiki’s first concrete data point on agentic commerce at 100M+ DAU scale outside the demo / proof-of-concept stage.

6. The four lessons from China’s “no-bullshit” approach. Ognibeni distils his argument into four operating principles that double as a diagnostic for Western AI initiatives:

   Lesson	What it counters in Western practice
   Aim for growth and revenue	The default cost-savings / fire-the-service-team framing — “Klarna fired all their service people two years ago because AI can handle that. Six months later they hired them back because the pilot worked great but edge cases make it extremely difficult to actually do this.”
   Go beyond demos	The demo-to-deployment chasm — “BCG, PwC and McKinsey all say the same: every company does something with AI, but very rarely we see real economic value coming out of this at scale.”
   Think in ecosystems	Western silo problem — “Chinese companies always think in ecosystems, we think in silos and these silos are not interconnected. That’s a real problem.”
   Optimize for trust	The reason BCG/PwC/McKinsey can’t find the economic value — “Very often the trust is lacking, and you can’t scale users if people don’t trust the system.”

7. The cliffhanger: search-driven e-commerce is the format AI agents kill. Ognibeni’s closing tease: “Here we very often only see search-driven e-commerce — and that will be handled by AI agents. Nobody will show up in your store when you only do search-driven e-commerce.” The implication: incumbents whose acquisition funnel depends on a customer typing a search query are the first wave to be disintermediated by an agent that already knows what the customer wants. This is the same disintermediation thesis Scott Price (DFI) names as ‘what keeps me up at night’ — the two videos hold the same thesis from opposite sides of the buyer/seller table.

What was actually ingested

Full ~28-minute keynote transcript via the youtube-transcript-skill (English auto-generated ASR). 664 timestamped segments. Includes the audience Q&A (~21:00–28:22). The talk is not chapter-segmented on YouTube; the wiki preserves the natural arc: ~0:00–4:30 China-as-time-machine framing + ByteDance; ~4:30–8:30 West-chases-AGI vs China-does-diffusion + Klarna anti-case; ~8:30–11:00 Xiaomi SU7 / dark factory; ~11:00–17:00 the four agentic-commerce cases (JD Joy Streamer, AQ, Qen one-sentence-purchase, Alibaba ecosystem); ~17:00–21:00 the four-lessons distillation + closing wave-of-Chinese-entrants warning; ~21:00–28:22 audience Q&A (sustainability of Shein-style throughput, Europe’s response to silos).

ASR notes:

• Speaker name “Bonibini” in opening line is ASR — the correct rendering is Björn Ognibeni (per YouTube channel description and visible on-screen lower-third). Wiki body uses Ognibeni throughout.
• Product names: “Cance AI” in ASR is Seedance (ByteDance’s text-to-video model); “AQ” / “AO” / “ACU” are all ASR variants of AQ (the Alibaba supply-chain chatbot, sometimes written Accio); “Qen” / “Quan” / “Q1” are ASR for Qwen (Alibaba’s LLM family, here referring to the Qwen-powered Alibaba agent product). The wiki preserves AQ and Qen as Ognibeni pronounced them in body prose, with the canonical name noted parenthetically on first use.
• “Shien” / “Sheen” / “Chien” → Shein throughout (the Chinese fast-fashion firm, here cited as the AI-first-platform exemplar).
• “Teu” / “Temo” → Temu throughout.
• “BYDongs” → BYDs (the Chinese EV maker BYD).
• “Bite Dance” / “Bance” / “Mike Dance” → ByteDance throughout.
• “Müster” / “Hoou” → Münster / Hangzhou respectively (geographic ASR drift).
• “Mckenzie” → McKinsey when the speaker means the consultancy.
• “Yan Lau” → Yann LeCun (the speaker references LeCun’s “LLMs are a dead end / world models are the future” position in the Q&A).
• Filler um/uh preserved in raw, suppressed in body prose where it interrupts reading.

Convergence with the wiki corpus

Source	What this video adds relative to it
MGI Race Takes Off (PDF)	MGI describes the omniscaler category and the AI foundation cluster structurally. Ognibeni names the specific Chinese AI-platform-firm cases the MGI PDF mostly treats as a sidebar: ByteDance’s tokens/GMV scale, Alibaba’s agentic-commerce stack (AQ + Qen + tokenized mandates), JD.com’s deployed virtual-host revenue, Xiaomi’s dark-factory production line. First wiki source with concrete RMB revenue lines attributable to Chinese agentic commerce.
MGI virtual event	The MGI panel defends omniscalers from inside the McKinsey-with-data frame. Ognibeni defends the same phenomenon from the outside-in European-practitioner-frustrated-with-EU-incumbents frame. Both surface the founder-control / long-horizon-bet / cross-unit-synergy triad — Ognibeni adds the “China is the only ecosystem not influenced by Silicon Valley” geographic premise as the explanation for why Europe keeps getting blindsided.
Spiegel — software is not a moat	Spiegel argues that the moat is the network-of-relationships, not the software. Ognibeni operationalises that with four lessons (growth-not-savings, beyond-demos, ecosystems-not-silos, optimize-for-trust) that read as a diagnostic Spiegel could apply directly to which Western firms have a chance to build that moat in agentic commerce.
Ries — the force that destroys	Ries: efficiency-chasing destroys growth from within. Ognibeni: Western AI deployments default to efficiency/cost-savings (Klarna) while Chinese deployments default to growth/business-model-renewal (Shein, BYD, Xiaomi, AQ). The two arguments converge on the same diagnosis from opposite sides.
enterprise-ai-adoption	Adds the BCG / PwC / McKinsey adoption-gap diagnosis in operator-narrated form, and locates the gap in four specific failure modes (trust deficit, demo-vs-scale, silo-vs-ecosystem, cost-vs-growth) — a finer-grained breakdown than the “micro-productivity trap” framing currently on the concept page.
strategic-foresight	Ognibeni’s China-as-time-machine framing is a textbook digital-scouting practice for digital transformation — “if you have different experiences, that’s where you find different ideas” — and complements MGI’s arena-creation-potion radar approach with a concrete geographic vantage.
ai-agents	The Qen one-sentence-purchase agent (100M+ users) and JD’s Joy Streamer (~$250M Double 11 sales) are deployed-at-revenue-scale agentic systems — significantly past the intern-entity / supervised proof-of-concept stage the concept page currently anchors on. Net-new evidence for the agents-in-production-at-consumer-scale claim.

Linked entities and concepts

• enterprise-ai-adoption — extended with the four-lessons diagnostic + Chinese agentic-commerce revenue lines.
• strategic-foresight — extended with the China-as-time-machine digital-scouting frame.
• ai-agents — extended with the agentic-commerce-at-scale exemplars (JD Joy Streamer, Qen one-sentence-purchase, AQ supply-chain builder).
• Björn Ognibeni — speaker; Hamburg-based practical visionary; co-founder ChinaBriefs.io + XRLab@MCM (University of Münster); UC Davis lecturer (“Rethinking Digital”). First wiki mention; deferred per the second-source promotion rule.
• E-commerce Berlin Expo — venue / publisher channel. First wiki mention; deferred.

Dangling (single-source mentions, deferred per second-source promotion rule): Björn Ognibeni, E-commerce Berlin Expo, ByteDance, Alibaba, JD.com, Xiaomi, Shein, Temu, BYD, Klarna, ChinaBriefs.io, XRLab@MCM, UC Davis, Seedance, AQ (Accio), Qwen, Fliggy, Yann LeCun. Promote on any second-source mention — Yann LeCun is the most likely to recur given his public position on LLM dead-ends; AQ / Qwen / agentic-commerce-as-product-category are likely to recur in subsequent MGI / DFI / agentic-AI ingests.

Open questions raised

• Are the JD Joy Streamer / Qen one-sentence-purchase revenue lines verifiable outside Ognibeni’s slide? The wiki holds them on the speaker’s authority pending an independent source (JD/Alibaba IR filing, an analyst note, a CNBC / FT report). Worth a lint flag if no second source surfaces within a quarter.
• Will the “search-driven e-commerce gets killed by agents first” thesis hold? Ognibeni’s closing tease and Scott Price’s “agentic-AI-disrupts-the-retailer-customer-relationship” worry (DFI interview) are the same prediction from buyer-side and seller-side respectively — track which Western retailers actually pivot their acquisition funnel away from query-typing within 18 months.
• Will Alibaba’s translation-guarantee model become the trust-scaling pattern other agentic-commerce platforms adopt? The “if the translation is wrong, it’s Alibaba’s fault” legal-liability shift is the trust-scaling innovation Ognibeni names as load-bearing. Watch whether Western marketplace platforms (Amazon, eBay, Shopify) adopt symmetric liability-shifts for their agentic offerings.
• What is the “world models, not LLMs” opportunity Ognibeni hands to Europe in the Q&A? Ognibeni cites Yann LeCun’s position that LLMs are a probabilistic dead-end for embodied AI (robotics, autonomous vehicles) and that world models are the next frontier — “all the basic research that LLMs are based on already came from Europe.” A future thread question: does Europe have the institutional capacity to specialise in the non-LLM next-generation architecture, or will the same Silicon-Valley-influence pattern repeat?

Related sources

• 2026-05-14-price-dfi-retail-asia-reinventing-how-it-sells — the same agentic-commerce-disruption thesis from the incumbent retailer’s vantage (DFI / Scott Price). Read both together as a buyer-side / seller-side pair.
• 2026-03-25-russell-bradley-mgi-race-takes-off-next-big-arenas — the macro-data MGI report that names the omniscaler category Ognibeni populates with specific Chinese cases.
• 2026-05-12-mgi-virtual-event-race-takes-off-next-big-arenas — MGI panel’s omniscaler defence; Ognibeni’s outside-in European angle complements Sastry’s built-recipe-vs-in-construction-recipe framing.
• 2026-04-26-how-to-win-when-software-is-not-a-moat-evan-spiegel-snapchat-ceo — Spiegel’s moat-is-network-not-software thesis; Ognibeni operationalises it.
• 2026-05-10-ries-lennys-force-destroys-companies-within — Ries on the internal force; Ognibeni on the symmetric external symptom.