This article is the 1062nd original work by DeepAtom.

"

AI Office Becomes the Next Major Battleground

By Meng Fanle | Author

Edited by DeepAtom Studio

The battles never end; they just keep going.

Over the past two decades, Baidu, Alibaba, and Tencent—the BAT trio—have engaged in fierce competition across e-commerce, O2O, transportation, payments, short videos, large models, AI entry points, and more. Nearly every sector capable of reshaping the industry landscape has seen these three companies collide head-on.

By the summer of 2026, the battleground for the three giants had shifted to AI office, bringing back that familiar flavor: Tencent, Alibaba, and Baidu all aimed their guns at the same target almost simultaneously, vying for the next-generation AI office entry point. Just during the Spring Festival, they were still locked in bloody battles over Chatbots.

The change first began in the U.S. market, where Anthropic took the lead in shaping the concept of 'work' into a widely accepted product form. In May 2025, Anthropic quietly launched a product: Claude Code. It was an AI agent capable of directly reading and writing to code repositories, executing commands, and submitting pull requests (PRs). Overnight, it swept through the entire programming industry. It then rapidly expanded into non-programmer work domains, effectively defining the product form of AI office.

For the first time, the industry reached a consensus on AI applications: Work-type products hold greater commercial value than Chatbots. At least for now, they are the products that can generate substantial profits. This is because they prove one thing: once agents truly take over workflows, users are willing to pay generously for results.

In February 2026, Anthropic's annualized revenue run rate (ARR) reached $2.5 billion. It then quickly ascended along a steep growth curve. Recent reports suggest its ARR will surpass $100 billion by the end of August. OpenAI is close behind, expected to reach $100 billion in ARR by year-end without issue. These are astonishing growth figures. For comparison, TikTok, the world's most powerful money-printing machine, is projected to generate $186 billion in revenue in 2025.

So, who can remain indifferent in the face of such a trillion-dollar market?

July 2026 marked a turning point: AI office was no longer a scattered experiment across various agent product lines but a full-scale, coordinated effort.

China's AI Work-type products actually began with 'lobster farming.' Tencent's WorkBuddy and Baidu Dazi started as lobster-class products before slowly iterating toward a Work-product form like Codex.

By July, as everyone could see, the three giants adjusted their product priorities: Tencent pushed WorkBuddy to the forefront, jokingly dubbed 'Tencent's new crown prince.' Alibaba consolidated capabilities like QoderWork, Wukong, and MuleRun into Qianwen Office, while Baidu formed an AI office matrix with products like Baidu Dazi (Dumate), GenFlow, and MiaoDa.

Tencent, Alibaba, and Baidu are all pursuing the same goal but starting from different positions due to their inherent differences.

Three Starting Positions

The previous generation of office application battles lasted nearly a decade. DingTalk, WeCom, Feishu, Baidu Wenku, and cloud storage all achieved their respective victories.

The advent of AI has reset the game. The core of the previous generation of office platforms revolved around instant communication, organizational collaboration, approvals, and documents, with the key focus on 'how people communicate and how organizations operate.' However, in the age of agents, the core logic of office software has shifted from 'communicate first, then execute' to 'deliver results first, then review and align.'

When agents can break down tasks, call upon tools, generate solutions, and complete initial execution, IM (instant messaging) is no longer the starting point of work. When agents can independently complete complex tasks in a closed loop, organizational collaboration becomes less critical.

This is where BAT diverge in AI office.

Tencent's WorkBuddy path starts from instant communication and social chains, evolving toward agents. WorkBuddy's advantage lies in the relationship chains of WeChat and WeCom. WeCom plays a pivotal role in connecting enterprises, employees, and external customers, especially in sales, customer service, and private domain operations, where it remains an irreplaceable entry point. However, in the age of agents, the core of an agent is task completion, relying not on relationship chains themselves but on task context. Merely having the ability to 'connect people' may not suffice in the AI office era. We've already seen Tencent enhance WorkBuddy's capabilities around 'tasks.'

Alibaba's Qianwen Office path starts from organizations and evolves toward agents. DingTalk's years of accumulated organizational relationships, approval workflows, and enterprise clients are Alibaba's most valuable assets. The value of Qianwen Office lies first in combining Tongyi model capabilities with DingTalk's organizational context. However, Alibaba has realized that DingTalk belongs to the previous generation of office products. Qianwen Office must not merely add an AI layer to DingTalk but become a more native agent product, prioritizing 'tasks first, rather than organizational collaboration first.'

Baidu Dazi is an office agent closer to AI-native principles. It first assumes the user wants to complete a task, then works backward to determine which files, tools, and data the agent needs to call upon. Baidu's approach to AI office starts from data and agent execution, not being as deeply tied to the previous generation of office entry points, allowing it to rebuild the AI office entry point more thoroughly according to AI-native logic. The entry point for AI-native office may not be 'initiate communication' but 'initiate a task.' Directly let the agent read materials, organize solutions, generate PPTs, build small tools, and then have humans review and modify. This logic naturally aligns closer to data assets and desktop-level execution capabilities.

Tencent excels in 'connecting people,' Alibaba in 'connecting organizations,' and Baidu bets on 'connecting data.' All three have chosen their flagship products to lead the charge, but Baidu Dazi is the only one initially positioned as a 'general-purpose agent + data foundation'—not dependent on any existing product ecosystem, unlike WorkBuddy built on WeChat or Qianwen Office built on DingTalk. It comes with its own data assets and execution capabilities. Keep in mind, Baidu Netdisk has 1 billion users and 100 billion GB of work files.

In the age of agents, whoever possesses the most authentic user work files and data precipitate (data accumulation), and whose general-purpose agent can truly execute, holds the deepest defensive moat.

What Cards Does Baidu Dazi Play?

Baidu Dazi made its high-profile debut at the 2026 Create Conference in May. Baidu founder Robin Li first proposed a new metric for the agent era—DAA (Daily Active Agents), emphasizing that AI's value should shift from 'Token consumption' to 'actual delivered results.'

Baidu Dazi was born from this philosophy. Its initial positioning was as a general-purpose agent that truly 'gets things done.' Baidu Dazi integrates core product capabilities like Baidu AI Search and Baidu Baike into built-in skills that can be called upon anytime, while strengthening long-term task execution and proactive decision-making abilities, becoming the unified entry point for users to enter the Agent world.

Baidu doesn't want users to switch back and forth between multiple AI products but hopes to consolidate actions like 'finding materials, summarizing, writing reports, creating PPTs, and building tools' into a single execution interface. For enterprise users, the value of this unified entry point is tangible: it reduces switching costs and improves implementation efficiency.

Today, in the AI office Track ( Track means 'track' or 'arena'), Baidu's differentiated strategy is a coordinated group effort. Baidu Dazi collaborates with products like GenFlow and MiaoDa, forming Baidu's proprietary AI office product matrix with unique advantages.

GenFlow already has 100 million monthly active users, excelling in long-document refinement, report writing, and video editing based on netdisk assets. These scenario-based capabilities provide unique support for Baidu's competition in the AI office domain.

Then there's MiaoDa, which already boasts over 35 million users and 3.5 million+ application generations. It offers zero-threshold no-code development capabilities, enabling agents to generate micro-applications and lightweight tools in real-time to meet personalized, cross-system long-tail business needs and complete deliveries.

Beyond that, Baidu Dazi's product rhythm is extremely fast, with daily updates, showcasing the agility of an AI-native team. It remains open, integrating with mainstream office tools like WeChat, Feishu, and Ruliu. More importantly, Baidu has prioritized enterprise-grade security, including secure sandboxes and folder-level permission controls. These capabilities directly address enterprises' biggest fears: what AI can see, modify, and whether it can access sensitive files. Baidu Dazi indeed has its own distinct strengths and advantages in security and trust.

In this light, Baidu is indeed entering the fray with a brand-new competitive logic: as AI office competition shifts from 'who has the bigger entry point' to 'who has deeper data and whose agents can get more done.'

Camp Divisions: The Big Tech Game

Today, China's AI landscape has formed two major camps: one consists of tech giants, represented by BAT; the other comprises entrepreneurial newcomers like Deepseek, Moonshot AI, and Zhipu AI. Tech giants pursue both foundational models and applications, while entrepreneurial newcomers continue to focus on models, striving to push the boundaries of intelligence.

Of course, AI office has never been a contest of single-point model capabilities. What truly determines victory are four dimensions: model agent capabilities, data assets, ecological toolchains, and B2B delivery—all must work in tandem. Models determine entry eligibility, data determines task completion, toolchains determine stickiness, and delivery determines monetization capabilities. Without any one link, the loop fails.

From the model side, all three have their aces. ERNIE Bot 5.1 ranks first domestically on the LMArena search leaderboard, Tencent Hunyuan is strengthening group-level RL infrastructure, and Alibaba Tongyi continues to double down on Coding and Cowork directions.

From the data side, the differences are even more pronounced. WeChat and WeCom hold communication and social relationship data, DingTalk holds organizational relationship data, while Baidu holds work file data. Once AI office enters the task execution phase, the most valuable assets are those files that can be directly called upon, edited, reorganized, and reproduced.

From the ecological side, Tencent integrates documents, netdisk, Lexiang, WeChat Pay, and ima; Alibaba relies on DingTalk and Alibaba Cloud's enterprise links; Baidu Dazi is more open, already connecting with WeChat, Feishu, and Ruliu. Whichever has a more complete tool and ecosystem is more likely to turn one-time trials into high-frequency habits.

From the delivery side, Alibaba, Baidu, and Tencent all have strong B2B sales and delivery teams. However, AI office demands different capabilities. How to redesign delivery processes and rebuild delivery capabilities based on user and client needs is a common challenge for all three.

Big tech may lack the agility of startups but excels at matrix warfare. BAT happens to be the three companies that have proven their comprehensive strength in the internet era—they have technology, data, ecosystems, and money. Moreover, they are highly determined.

No Quick Victories for Anyone

BAT dominated the internet era with search, e-commerce, and social media, each occupying a pole for years, ultimately reaching a stalemate.

However, the AI office battle features shorter fronts, faster rhythms, and more dense ( dense means 'frequent') product iterations, significantly intensifying competition. Yet, the duration of competition may lengthen, with no side able to pursue a quick victory.

Robin Li shifted the competitive focus from model capabilities to execution capabilities, based on his consistent belief in application-driven progress: Tokens represent costs, not revenue; what truly matters are agents capable of delivering results. This judgment is especially crucial for Baidu, as Baidu Dazi aims to become the key entry point for task execution.

Placed on a larger timescale, the significance of BAT's reunion may not merely be about which products win or lose but whether Chinese enterprises' work methods over the next decade will be redefined. Whoever first integrates data, agent capabilities, toolchains, and delivery into a cohesive whole stands a better chance of transforming 'AI office' from a Track ( Track means 'track' or 'arena') into a generation of productivity infrastructure.

If Baidu Dazi can push 'deepest data, strongest agent, fastest iteration' to their limits and translate them into 'most enterprise clients, best implementation, strongest mindshare,' it won't just be flying a product flag but repositioning Baidu in this AI office battle.

The window of opportunity for AI office is closing. Everyone is striving to make clients, employees, and partners default to: for work, one should first open this entry point.