Why China Is Built Perfectly For AI Agents

Vídeo em Inglês, publicado pelo canal BetterWay.

About this content: The summary, key points and topics below are generated automatically from the video transcript. They reflect the content, opinions, claims and predictions presented in the video and do not represent a position, validation or fact-check by IA Dots.

Video summary

The video discusses the stark contrast between the booming AI agent startup ecosystem in the United States and the dominance of tech giants in China. It explains how America's fragmented digital web drives independent startups to build new infrastructure layers, while China's integrated super apps allow platform-owned agents to operate seamlessly within closed ecosystems.

Key points

• The US leads in agentic automation startups, while China's usage of open-source tools is high, but development is concentrated among a few giant companies.
• The American digital landscape is fragmented, featuring isolated websites with barriers like captchas and individual logins, which creates a market opportunity for independent AI agent startups.
• China's digital life is centralized around super apps like WeChat and Alipay, allowing platform-native AI agents to integrate identity, payments, and services under one roof.
• Chinese tech giants restrict external AI agents that bypass their apps' ads and commissions, making an agent's utility dependent on whether the platform permits it.
• US financial institutions and startups are building a new infrastructure layer—such as agentic tokens, specialized wallets, and verification protocols—to bridge the gap in the open web.

Topics

AI agentsStartup ecosystemsUS vs China techSuper appsDigital fragmentationPayment protocolsWeb infrastructure

Video transcript

Transcript from available captions; it may contain errors.

Have you ever wondered why does the US have so many different AI agent startups but you don't see many from China? Startups like Cognition's Devon works as an autonomous software engineer. Decagon builds AI concieres for brands like Avis and Deutsche Telecom. And Skyfire gives agents their own identity and payments. And there's so many more. If you look at the numbers across AI agents, reports show that 325 agentic automation startups are in the United States with India second and Germany third. But China doesn't even make the top three. This gap is not because Chinese people don't use agents. They do and in huge numbers. China's usage of openclaw has even passed America's according to security scorecard. The difference is who builds them and how the two countries ecosystems are built in the first place. First, the ecosystem. Picture an AI agent trying to buy you a pair of shoes on the American internet. Every step is a wall. The website first asks it to prove you are human. A capture can't tell the difference between your assistant and a scraper, so it blocks both. Then comes the login with its own identity check. Then the checkout built for a person with a browser and a credit card, not an AI with credentials. None of these systems talk to each other as well. Every site runs its own identity, its own payments, and its own rules. They were all designed on one assumption. A human is on the other end. Some sites go further and fight back. Walmart's checkout asks visitors to click a button to prove they're human. If an agent interrupts it, the check fails and the agent gets booted out. Yelp bars non-human traffic unless the agent pays for a data license. Even Amazon has most recently blocked Meta's Muse agents and Perplexity's agents from purchasing from its platform and is even fighting Perplexity in court for this, arguing that agents threatened their advertising business because it can't tell which ads humans saw and which went to bots. That fragmentation is exactly why agent startups are booming in the US. Americans have scattered their digital lives across dozens of separate accounts. Your bank, airline, grocery store, and email each have their own login and their own system, and no single company owns them all. But most of it can still be reached through a browser. So independent startups can build agents that work across all of it. China's story is the opposite. In China, daily life already runs through a handful of super apps. People order food, hail taxis, pay bills, and renew official documents all inside them. The AI agents released by the few giant companies that run all of China's ecosystem are the only ones needed to do everything. Mobile payment is the clearest example. One Brookings analysis found that more than 92% of China's mobile payments ran over just two platforms, Alip Pay and WeChat Pay. Identity, payment, and services already sit under one roof, so there's nothing to stitch together. So, Chinese agents plug straight in. Alibaba's Quinn app can reach the catalog of Pao and T- Mall, Alibaba's two largest shopping marketplaces with more than 4 billion items and Alipe checkout built in. When Alibaba handed out free bubble tea vouchers that worked only through Quen, users placed more than 10 million orders in 9 hours. Even the bills. Bite Danc's DBA now lets you pay water, electricity, and gas bills by voice or text. speak to it and it pulls up the payment page. Even DeepSeek's model was added to WeChat and Alipay. That's the idea of an agent inside an ecosystem that already holds your accounts to everything. Still note that DBA hands you over to Doí's bill page and you still bind your account and confirm the payment yourself. Even in China, agents don't pay on their own yet. Alipe still asks for manual confirmation and WeChat limits its agent to a locked spending quota. Then the race between the giants start. In one week in June, JD published a payment protocol for agents. Alipe began testing an AI version of its app and WeChat Pay launched an AI only card. One Chinese outlet summed up the logic. Whoever locks down the rules first collects the tolls. But this is also the catch. China's ecosystem is integrated, but it's also closed. In December 2025, Bite Dance released a test phone whose agent could use any app without the app's permission. It read the screen and tap the buttons like a person. Within about a day, WeChat, Alibaba's apps, and Muan restricted it. Taoba and several banks stopped it from logging in or paying. Their reasons were simple. The agent skipped the ads, the commissions, and the time users spend inside their apps. Tencent's chairman called the screen sharing extremely unsafe. That's why Tobal with 382 million monthly users is one of China's most used AI apps. Yet, public sources show it completing only one of 10 everyday tasks tracked. Agents with half its audience or less can order food, book travel, and pay bills because their parent companies own those services. In China, an agent isn't useful because it's smart. It's useful because the platform lets it in. Now, look back at America. China's answer to fragmentation is that a few giant platforms already own everything. America has no giant to lean on, so it's trying to solve the problem from scratch. A whole new layer of infrastructure built just for agents. Visa launched its trusted agent protocol after AIdriven traffic to US retail sites surged more than 4,700% in a year. It lets merchants verify an agent and tell it apart from a malicious bot. Mastercard registers agents and issues them special agentic tokens. Stripe built shared payment tokens so an agent can pay without ever seeing your card number. And then startups fill in the rest of the areas. Skyfire builds identity for agents. Rye lets an agent complete a purchase on almost any online store. And Cina Labs is building an AI native bank. One market map counted 334 products that help agents move money. Agent browsers, agent ready websites, agent wallets, and agent cards are all products of the same gap in the US ecosystem. And demand keeps climbing. Meta's Muse agent was downloaded 5 million times in its first month and Agentic web traffic tripled in a single month. Shopify is already giving agents their own checkout lanes as well. The mess of the open web has become a startup opportunity. But that layer only works if websites choose to let agents through the door. That's why the two countries look so different. US companies like OpenAI and Meta build the assistant first, then send it out to other companies sites. In China, Tencent Shiawei lives inside WeChat and works with businesses already on the app. The numbers show this as well. AI features embedded inside Super Apps reach 706 million users in China against 287 million for standalone AI apps. So the US has an open but fragmented web. Its problem is technical and the fix is a new layer built from the outside by startups. China has an integrated but closed one. Its problem is political and the winners are the platforms that own the services. Neither side has settled it yet though. MIT Technology Review notes that the rules for how agents deal with websites are still being written and whether businesses can block agents will decide if the web works for them or against them. So, China's ecosystem is perfectly built for how agents need to operate as everything is already connected and it's especially beneficial for the agents that are owned by those platforms. While the US with its new protocols and agent passports may be heading toward the fight over who decides which agents get in.

Assistir ao vídeo original no YouTube

Assuntos: Agentes de IA, Assistente pessoal, Muse. Publicado no YouTube em 2026-10-26.

Seleção automática por assunto e idioma. O vídeo pertence ao canal de origem.