LAIKA 

 2026/10/15 

New Playbooks and Old Challenges Coexist.

The personal AI agent landscape is entering a new phase.

A month ago, Meta launched Muse, a personal AI agent that topped the free iOS charts in the U.S. and Canada. Chinese tech giants swiftly responded: ByteDance accelerated its “Spell” personal assistant project, Tencent launched LightVela, Alibaba’s QianWen prepared “QianWen Go!,” Ant Group’s “LingGuang” began limited beta testing against Muse, and unicorn startup Manus introduced Cue, an independent personal AI agent app.

Unlike the “Lobster Agent” craze sparked by OpenClaw earlier this year, this wave of personal AI agents is more concentrated: tech giants and leading startups dominate, shifting the battleground from B2B AI office tools to C-end consumer scenarios.

With this year’s shift in AI investment logic, the market no longer solely competes on model parameters or computational arms races. Commercialization efficiency has become the top priority. Tech giants are overhauling their AI product lines, transitioning from “launch first, find scenarios later” to “design closed loops first, then deploy products.”

Muse has unveiled not just a global race for personal AI agents but a Refactoring (reconstruction) of AI business models.

01

From B2B to C2C: Tech Giants Pivot En Masse

Over the past two years, tech giants focused their AI efforts on B2B, yielding early results.

Products like Tencent’s WorkBuddy, ByteDance’s Doubao Work, Alibaba’s QianWen Office, and Baidu’s Dazi shared a common trait: they assisted “workers” and charged “employers.”

But Muse changed everything. Targeting consumers directly, it handles high-frequency life scenarios like shopping, travel, and scheduling. By upgrading AI from a “Q&A tool” to a “task agent,” it became a “digital service” capable of charging users.

This sparked a domestic frenzy among tech giants to accelerate personal AI agent deployment. ByteDance, for instance, ramped up R&D on “Spell” before National Day, a project initiated in April and led by the Doubao mobile assistant team. The goal is to integrate “Spell” with Doubao for a swift launch.

Tencent followed suit on multiple fronts. In late September, it launched LightVela, a “cloud-based personal assistant” accessible via WeChat, handling daily tasks 24/7—though still in developer preview. Meanwhile, Tencent’s Yuanbao is evolving into a personal AI agent. At the 2026 Cloud AI Industry Summit, Tencent unveiled over 20 agents under the “Efficiency Agent Toolkit,” covering personal, office, and enterprise scenarios.

Additionally, Tencent’s “Xiaowei,” a native AI assistant tested since June, supports tasks like food delivery, package tracking, and ride-hailing, as well as summarizing recent WeChat moments or chat logs.

Alibaba announced at the 2026 Cloud Town Conference on September 22 that QianWen would accelerate its Personal Agent development. Rumors suggest an imminent launch of “QianWen Go!,” an agent embedded in the QianWen app to enhance task-handling capabilities, such as negotiating flight delay compensation, scheduling car maintenance, or requesting Taobao price protection refunds. Meanwhile, Ant Group’s “LingGuang” began a new round of limited beta testing, benchmarked against “Meta Muse.”

Beyond tech giants, renowned unicorn Manus—after a rollercoaster journey of product hype, headquarters relocation, company sale, acquisition termination, and full return to the domestic market—recently secured $500 million in new funding. It launched Cue, a smart assistant for personal life scenarios, supporting independent email, phone numbers, wallets, and dedicated cloud PCs for agents. Cue enables multi-agent collaboration and 24/7 personal task management.

AI agents are undoubtedly this year’s market focal point. Earlier, OpenClaw triggered a “Lobster Agent” boom, with independent developers and small teams rapidly building agent demos. While diverse, most were limited to web and PC tool integrations, a grassroots celebration for developers and geeks. This time, players wield complete resources: large models, account ecosystems, payment networks, and terminal hardware.

This signifies a shift from a “wild west” era to a “professionalized” battlefield. After B2B AI office agents sparked a “race for the office,” tech giants are now targeting consumers directly.

02

New Playbooks, Old Challenges for AI Agents

The logic behind personal AI agents differs fundamentally from conversational AI. Standard large models operate on a question-and-answer basis, merely outputting information. Personal AI agents, however, hold executive power: they read screens, jump across apps, fill forms, initiate payments, and retain long-term personal preferences.

Meta’s Muse, for instance, assists with online shopping, email management, and other tasks upon user authorization. Its chat-like interface lowers entry barriers and supports customization, such as personalized names. Muse offers a free tier and two subscription plans at $20 and $100 monthly. This reflects Meta’s AI strategy: simplicity drives daily integration, encouraging users to pay for premium services.

While all tech giants and leading startups launch uniquely featured personal AI agents, two core contradictions persist.

First, security and privacy risks. During internal testing, Muse exposed multiple safety and privacy flaws, such as unauthorized password changes and reading private messages without consent. Post-launch, Amazon blocked Muse from shopping on its site, citing privacy and security concerns over credential handling.

Meta claims Muse incorporates multi-layered safeguards, requiring user confirmation for emails or purchases, while low-risk tasks run automatically. Each Muse operates on an isolated “cloud-based dedicated computer,” inaccessible to other agents. A separate supervisor agent monitors Muse’s actions.

Second, breaking ecosystem barriers for cross-platform operations. When Doubao Mobile launched in December, it showcased an “ideal state”: AI could execute cross-app tasks, like ordering food via Meituan or price-comparing on Taobao. But reality struck: WeChat triggered security alerts for “abnormal login environments,” bank apps demanded AI assistant shutdowns, and Alipay and major banks imposed blocks.

Doubao Mobile’s tragedy highlighted a power struggle between “agents” and “super apps.” On October 14, Doubao Mobile released a “blacklist” of 150+ apps, including WeChat, Taobao, Meituan, Xiaohongshu, and DeepSeek, that explicitly reject agent automation on their screens.

Muse currently integrates with third-party services like Google Workspace, Ticketmaster, and Spotify via public APIs. However, Chinese super apps like WeChat, Taobao, and Meituan have no obligation to open APIs to rival agents.

Two product types may emerge: “closed-loop ecosystem agents” (e.g., QianWen Go! leveraging Alibaba’s e-commerce and open platform, Xiaowei relying on WeChat mini-programs and services, Spell integrating ByteDance’s content and local life) offer seamless but bounded experiences; and “protocol-stitched agents” (e.g., Cue and LightVela) use MCPs and messaging gateways for cross-platform scheduling, offering freedom but facing constant restrictions.

Objectively, without systemic solutions to privacy and ecosystem barriers, even the coolest personal AI agents may remain niche, appealing only to early adopters in limited scenarios. Only by addressing trusted execution and cross-ecosystem collaboration can they evolve from toys into infrastructure.

03

Tech Giants’ Shifting AI Product Strategies

Beyond the personal AI agent hype, a notable shift occurs in how tech giants approach AI products: the focus has moved from pure technical competition to product delivery, user value, and commercial closed loop (closed loops).

Over the past two years, domestic tech giants prioritized scaling large models, launching numerous AI products quickly, and aggressively marketing to capture market share. This year, investment focus shifted from “how strong is the tech” to “how does AI make money.” Firms are doubling down on AI compute while streamlining organizations and product lines, eliminate (phasing out) projects without clear monetization paths. Mergers, shutdowns, and upgrades are now commonplace.

For example, Alibaba merged QoderWork, Wukong, and MuleRun into “QianWen Office” to deeply integrate with DingTalk. ByteDance folded Feishu’s team into Doubao, launching “Doubao Work,” while rebranding the agent creation platform “Kouzi” as “Kouzi Programming.” Tencent concentrated resources on WorkBuddy, which surpassed 13 million DAUs in July, and recently shut down QClaw and Yuanqi. Baidu’s Wenxin Agent Platform closed agent creation, “clone,” and workflow entry points in August.

Meanwhile, Muse introduced a new AI commercialization narrative. Traditional AI products monetized via memberships or token sales, but Muse targets transaction fees from agent-brokered deals.

Analysts suggest Muse may pivot from subscriptions to transaction commissions, payments, and ads. While basic features remain free and advanced ones require subscriptions, the real potential lies in Meta taking a cut of purchases made through Muse as the “default shopping gateway” or charging merchants “agent referral fees.” This represents a new traffic monetization logic: selling not user attention but user “agency.”

This holds significant reference value for domestic players like BAT, ByteDance, and Manus. ByteDance could integrate agents into Douyin e-commerce and local life for consumption, exploration, and shopping. Tencent could leverage WeChat’s ecosystem, mini-programs, and payments for services and local transactions. Alibaba and Ant Group’s agents naturally fit e-commerce, claims processing, utility payments, and financial services. Manus Cue, as an independent personal agent, offers cross-platform task delegation and payments with greater flexibility.

The ultimate choice of business model—subscription-first, transaction commissions, or a hybrid of ads and transactions—will shape the future market landscape.

Conclusion

The eve of the personal AI agent battle has arrived. Muse proved C-end demand is real, and domestic tech giants have secured their initial positions. With high-frequency consumer marketing nodes like New Year and Spring Festival less than three months away, tech giants are not pursuing immediate mass monetization. Instead, they’re refining product experiences, perfecting ecosystem rules, and gathering user feedback to prepare for a full-scale C-end AI offensive in spring 2027.

Yet, the key to victory in this round lies not in model benchmarks or captivating agent narratives but in stably resolving privacy, cross-app execution, and commercial closed loops.

- END -