China experienced a surge in base models, the do-everything big language models that support the AI revolution last season. This time, the focus has shifted to AI officials —systems that are less about responding to customers ‘ questions and more about freely accomplishing stuff for them. &nbsp,

There are currently numerous Chinese companies creating these general-purpose online resources that can answer emails, browse the internet, and even create an interactive website. Many of these have emerged in just the past two decades, following in the footsteps of Manus—a basic AI agent that sparked week of social media frenzy for encourage codes after its limited-release start in early March. &nbsp,

These new AI agents are not actual huge language versions. Otherwise, they’re built on top of them, using a workflow-based construction designed to get things done. Many of these methods also introduce novel ways of interacting with AI. They are designed to manage and carry out multistep tasks, such as booking airlines, managing schedules, conducting research, using additional tools and remembering instructions, rather than just chatting back and forth with customers. &nbsp,

China might be the first to develop these kinds of providers. The country’s well-integrated app ecosystems, quick product cycles, and technologically proficient user base could create a conducive environment for incorporating AI into everyday life. &nbsp,

Because the best European models don’t run inside China’s firewalls, its leading AI broker startups are currently focusing their efforts on the international market. But that may change soon: tech giants like ByteDance and Tencent are developing their own AI providers that could extract information from their extensive ecosystem of programs, which predominate in some aspects of daily life in the nation. &nbsp,

A mix of ambitious businesses and established tech giants are currently testing how these tools may actually work in practice, and for whom, as the competition to determine what a valuable AI agent looks like progresses.

Set the standard

It’s been a whirlwind several times for Manus, which was developed by the Wuhan-based company Butterfly Impact. The business put the item on a bold world fair and hired lots of new people after raising$ 75 million in a cash large led by the US venture capital firm Benchmark. &nbsp,

Even before membership opened to the public in May, Manus had become a reference point for what a large, consumer‑oriented AI adviser should accomplish. This “general” broker is designed to be able to assist with daily tasks like getaway planning, stock comparison, or your child’s school project rather than just handling business specific tasks. &nbsp,

Unlike previous AI agencies, Manus uses a browser-based platform that lets customers manage the agent like an apprentice, watching in real time as it scrolls through online pages, reads articles, or codes actions. Additionally, it actively asks clarifying issues and provides environment for upcoming jobs.

” Manus represents a promising solution expertise for AI providers,” says Ang Li, cofounder and CEO of Simular, a startup based in Palo Alto, California, which creates computer use agencies, AI agents that control a digital system. ” I believe Chinese startups have a huge advantage when it comes to designing consumer products, thanks to cutthroat domestic competition that leads to fast execution and greater attention to product details”.

The competition is moving quickly in the case of Manus. For instance, Genspark and Flowith, two of the most buzzy follow-ups, already have benchmark scores that match or exceed Manus ‘. &nbsp,

Through what it calls multicomponent prompting, Genspark, led by former Baidu executives Kay Zhu and Eric Jing, connects numerous small” super agents” The agent performs tasks like creating slide decks or making phone calls, and can switch between several large language models. He also accepts both text and images. Whereas Manus relies heavily on Browser Use, a popular open-source product that lets agents operate a web browser in a virtual window like a human, Genspark directly integrates with a wide array of tools and APIs. The business, which was launched in April, claims to already have over 5 million users and generates more than$ 36 million in annual revenue.

Flowith, the product of a young team that first attracted national attention in April 2025 at a developer event held by the well-known social media app Xiaohongshu, adopts a different approach. Marketed as an “infinite agent”, it opens on a blank canvas where each question becomes a node on a branching map. Users can backtrack, create new branches, and store results in personal or sharable “knowledge gardens” —a design that resembles project management software ( think Notion ) more than a standard chat interface. Every task or inquiry creates its own mind-map-like graph, promoting more nonlinear and imaginative interaction with AI. Flowith‘s core agent, NEO, runs in the cloud and can perform scheduled tasks like sending emails and compiling files. The founders want the app to be a “knowledge marketbase,” and they want to “learn the secrets of AI knowledge from the only people who can make money from it.”

What they also have in common with Manus is a global ambition. Both Genspark and Flowith have stated that their primary focus is the international market.

A global address

Startups like Manus, Genspark, and Flowith, which were all founded by Chinese entrepreneurs, could seamlessly blend into the global tech scene and effectively compete abroad. Founders, investors, and analysts that MIT Technology Review has spoken to believe Chinese companies are moving fast, executing well, and quickly coming up with new products. &nbsp,

Money amplifies the desire to launch overseas. Customers there pay more, and there are plenty to go around. On a podcast, Manus cofounder Xiao Hong bragged,” You can price in USD, and with the exchange rate that’s a sevenfold multiplier. We’ll still make more money than we do in China, even if we’re only operating at 10 % power because of cultural differences abroad.

But creating the same functionality in China is a challenge. Due to geopolitical risks and difficulties with regulatory compliance, the majority of US AI companies, including OpenAI and Anthropic, have left China. Users turned to VPNs and third-party mirrors to access tools like ChatGPT and Claude after their absence initially sparked a black market. That vacuum has since been filled by a new wave of Chinese chatbots—DeepSeek, Doubao, Kimi—but the appetite for foreign models hasn’t gone away. &nbsp,

For instance, Manus employs Claude Sonnet from Anthropic, which is widely regarded as the best model for agentic tasks. Manus cofounder Zhang Tao has repeatedly praised Claude’s ability to juggle tools, remember contexts, and hold multi‑round conversations—all crucial for turning chatty software into an effective executive assistant.

However, the company’s use of Sonnet has made its agent in China ineffective without a VPN. You’ll see a notice stating that the team is “working on integrating Qwen’s model,” a special local version built on top of Alibaba’s open-source model, when you open Manus from a mainland IP address. &nbsp,

The absence of Claude Sonnet models “limits everything we do in China,” according to an engineer who spoke to MIT Technology Review in an attempt to avoid being arrested. He continued, DeepSeek’s open models still hallucinate too frequently and lack training in real-world workflows. Developers we spoke with rank Alibaba’s Qwen series as the best domestic alternative, yet most say that switching to Qwen knocks performance down a notch.

According to Jiaxin Pei, a postdoctoral researcher at Stanford’s Institute for Human Centered AI,” building agentic capabilities in base LLMs has become a key focus for many LLM builders, and it will only be a matter of time until people realize the value of this.

Manus is doubling down on the audiences it can already attract at this point. In a written response, the company said its “primary focus is overseas expansion”, noting that new offices in San Francisco, Singapore, and Tokyo have opened in the past month.

A super app approach

Although the idea of AI agents is still relatively new, China’s market for consumer-facing AI apps is already crowded with major tech players. DeepSeek remains the most widely used, while ByteDance’s Doubao and Moonshot’s Kimi have also become household names. Most of these apps, however, are still more focused on task execution and chat than entertainment. Although early versions of these agents remain inconsistent in quality and rough around the edges, China’s big tech firms have pushed them to roll out their own user-facing agents. &nbsp,

Coze Space, an AI agent based on its own Doubao model family, is being tested by ByteDance. Users can choose between “plan” and “execute” modes, allowing them to either direct direct control the agent’s actions or take a step back and watch it work independently. It connects the company’s own Lark office suite to 14 well-known apps, including Git Hub, Notion, and Notion. Early reviews say the tool can feel clunky and has a high failure rate, but it clearly aims to match what Manus offers.

In addition, Zhipu AI has developed a free agent based on its proprietary ChatGLM models, AutoGLM Rumination. Minimax Agent has been released by Shanghai-based Minimax. Both products look almost identical to Manus and demo basic tasks such as building a simple website, planning a trip, making a small Flash game, or running quick data analysis.

Large companies have plans to change that despite the limited usability of the majority of general AI agents introduced in China. Tencent CEO Liu Zhiping teased an agent during a May 15 earnings call who would integrate automation into China’s most widely used app, WeChat. &nbsp,

WeChat is already the home to millions of mini-programs that act like embedded apps, as well as messaging, mobile payments, news, and other services. These initiatives give Tencent, its developer, access to data from millions of services that abound in China’s daily life, an advantage that most rivals can only envy.

Historically, China’s consumer internet has splintered into competing walled gardens—share a Taobao link in WeChat and it resolves as plaintext, not a preview card. China’s tech giants have long resisted integration with one another, choosing to wage platform war at the expense of a seamless user experience, in contrast to the more interoperable Western internet.

However, the use of miniprogrammes has allowed WeChat to have unrivaled reach across all services, including those that used to be interoperable, like grocery orders and gym reservations. An agent able to roam that ecosystem could bypass the integration headaches dogging independent startups.

Alibaba, the e-commerce giant behind the Qwen model series, has been a force in China’s AI market, but it has been slower to release consumer-facing goods. Qwen didn’t have a dedicated chatbot app until early 2025, despite being the most popular open-source model on Hugging Face in 2024. In March, Alibaba rebranded its cloud storage and search app Quark into an all-in-one AI search tool. DeepResearch, Quark’s most agent-like effort to date, was available by the end of June. &nbsp,

ByteDance and Alibaba did not reply to MIT Technology Review‘s request for comments.

The most recent Chinese AI agents “reflect the all-in-one, super-app approach,” according to Li of Simular, who previously worked at Google DeepMind on AI-enabled work automation. ” In contrast, AI agents in the US are more focused on delivering services to particular verticals.

Pei, the researcher at Stanford, says that existing tech giants could have a huge advantage in bringing the vision of general AI agents to life—especially those with built-in integration across services. He claims that the market for customer-facing AI agents is still in its early stages, with a lot of issues involving liability and authentication. However, businesses that already have operations across a range of services have a logical advantage when deploying agents on a scale.