In the AI era, ByteDance also aims to win in B-side productivity and vertical technology scenarios.

On September 15, Liang Rubo, CEO of ByteDance, who had not made a public appearance for a long time, appeared at the "2026 Feishu Future Unlimited Conference & Doubao Work Kick-off Conference." His speech was not originally scheduled in the official agenda.

His last public appearance was in June when he spoke via video at the Volcano Engine FORCE Conference. This appearance marked his first attendance at the Feishu Future Unlimited Conference in years, highlighting ByteDance's emphasis on Feishu and Doubao Work.

Xie Xin, CEO of Feishu, remains highly influential within ByteDance. At the conference, Xie Xin released a new version of Feishu—Feishu 8.0, reconstructed for intelligent agents—after three years, along with the launch of China's first team intelligence product, "Doubao Work Partner," achieving native integration between Feishu and Doubao Work.

This marked the first complete public debut of ByteDance's AI To B business following an organizational restructuring in late July.

On September 16, the Bloomberg Billionaires Index showed that Zhang Yiming, the 43-year-old founder of ByteDance, had a net worth exceeding $105 billion, surpassing Indian tycoon Gautam Adani to become Asia's richest person.

Zhang Yiming's wealth growth is attributed to TikTok's global expansion and ByteDance's deep Layout (strategic Layout can be translated as "strategic Layout " or "deep investment") in AI, including popular products like the Doubao large model and video generation model Seedance. However, it cannot conceal another fact: ByteDance's accumulation in AI programming and To B productivity tools remains relatively weak. Liang Rubo's public support for Feishu and Doubao Work reflects ByteDance's ambition to win in B-side productivity and vertical technology scenarios in the AI era.

A WeChat Moment and a Failed Message Sent Ten Years Ago

Liang Rubo's speech began with several old screenshots.

The images showed the earliest design sketches of Feishu from early 2017, featuring only basic messaging functions without even a calendar. Beside them were two messages Xie Xin had sent him that year, followed by two red exclamation marks—indicating the messages failed to send, and not even the profile pictures were displayed. "Comparing it to Feishu today, I feel deeply emotional," Liang Rubo said.

Next, he pulled up a WeChat Moment he posted in April 2019. At the time, Feishu launched its global version, and he forwarded the post, writing, "This is the product I've been holding back the longest since joining ByteDance." While ByteDance's To C products typically take three months from inception to launch, Feishu took three years.

Liang Rubo shared these anecdotes not merely for nostalgia but to lead into a core judgment. "By the end of last year, the agent capabilities of large models began to reach a usable stage, and the implementation of AI in productivity scenarios accelerated significantly," Liang Rubo said. He believes that agents do not exist in isolation but need to collaborate with humans, the environment, and other agents, and that agent capabilities are closely tied to models—requiring tight integration between agents, models, and the collaborative environment to truly deliver value.

Based on this judgment, ByteDance completed an organizational restructuring two months ago, integrating the strengths of Doubao, Feishu, and Volcano Engine to focus on creating superior work and productivity products. Each business has a clear role and division of labor: Doubao Work provides intelligence, Feishu serves as the collaboration platform hosting enterprise contexts and tools, and Volcano Engine supplies models, computing power, and development tools. Thus, ByteDance has established a complete AI office value chain by connecting model capabilities, application scenarios, and infrastructure.

Liang Rubo also reiterated ByteDance's 2026 annual keyword—"Scaling New Heights"—marking his second public statement on the company's AI strategy. At the Volcano Engine FORCE Conference in June, he announced via video that "Scaling New Heights" would be the annual keyword, focusing on climbing the highest peak in the AI era.

He examined AI within the 50-year trajectory of the IT industry. Over the past 50 years, only four peaks have emerged in IT: PC, Web, Mobile, and AI. He deliberately instructed colleagues to draw the first three peaks half as tall "because, in comparison, the AI peak will be much higher."

More importantly, such peaks are rare, occurring every 15-20 years, making them precious opportunities. However, he then added a cautionary note to instill a sense of urgency in the team: "Even if we scale the last peak, we may fall when the new one emerges."

Office collaboration is a fundamental need reinvented with each technological shift. Liang Rubo cited Microsoft as an example: from Office to Office 365 to Teams, Office thrived through three technological transformations, even solidifying a stronger position than Windows. In his view, each technological shift redefines office work, and the AI-driven transformation will be even more profound, representing a qualitative leap.

"Feishu boarded the Mobile era train; we hope to fully seize its opportunities in the AI era," Liang Rubo said.

Feishu 8.0: Rebuilt for Agents

Over the past decade, collaborative office solutions addressed "how people work together"; entering the AI era, Feishu aims to solve "how people and agents work together." Xie Xin opened his speech by asserting that Feishu is the best collaborative office product for the AI era.

Feishu 8.0 is its first answer in the agent era.

Instead of starting with a feature list, Xie Xin made a memorable statement: "Feishu no longer serves only humans but also agents." He announced that "we've redesigned Feishu for agents" in Feishu 8.0.

In other words, Feishu 8.0 is not a routine version update but a systemic reconstruction to accommodate agents. The core goal is to enable agents to function "like humans" within Feishu. Thus, Feishu 8.0's three key upgrades all revolve around human-agent collaboration.

First, collaboration between humans and agents should be as natural and fluid as with other humans. In group chats, users can add agents by searching their names, just like adding human colleagues; users can invoke an agent's quick functions via "/" and view its thought process to understand how it derives answers; multiple agents can also collaborate autonomously to complete complex tasks.

Second, agents should be able to use Feishu's various products like humans. They can view group chat messages, reply, write documents, operate multi-dimensional tables, schedule tasks, join meetings, access cloud drives, and even initiate approval processes. Agents can perform nearly everything humans can do in Feishu.

Xie Xin mentioned a background detail often overlooked: long before OpenClaw's popularity, Feishu was one of the easiest chat platforms for lobster-like products (referring to agent products) to integrate with. Various agent products prioritized integrating with Feishu, and Feishu frequently appeared in agent tutorials and videos. Judging by GitHub stars, the Feishu CLI project's popularity leads peers by nearly sixfold.

This reflects Feishu's long-term investment in CLI capabilities. Since its March launch, Feishu has increased the number of functionalities open to agents via CLI from 247 to 767, improved call success rates from 78% to 95%, and accelerated overall call speeds by 39%. This means agents can now leverage nearly all of Feishu's features.

Xie Xin demonstrated an example: while humans might use Feishu's drawing board for simple flowcharts, agents can create highly complex, editable infographics. Agents have shown creativity surpassing humans in using certain tools.

Third, enterprise management of agents should be as clear, secure, and controllable as with humans. This is a top concern for enterprise clients: many worry that agents might overstep permissions and access information they shouldn't.

Feishu 8.0's solution is to keep an agent's permissions aligned with its user's. If an employee cannot see certain information, neither can their agent; enterprise administrators can uniformly manage agents, track usage, set AI usage caps, and specifically manage high-consumption models and scenarios.

Native Integration: "No Switching" Matters

The native integration of Feishu and Doubao Work is Feishu's most crucial recent update and one of the most underestimated aspects of this launch.

Both share the same enterprise account system, allowing Doubao Work to access a more complete enterprise context and enabling users to invoke Doubao Work anytime within Feishu to handle tasks. Xie Xin emphasized that Doubao Work is "not just an integrated agent or a third-party app requiring switching but is fused with Feishu at the system level."

This integration offers three core advantages: professionalism, business understanding, and compliance. Doubao Work supports generating and partially modifying Word, Excel, PPT, web pages, and apps, handling complex tasks through computer operations, multi-agent collaboration, and continuous cloud operation. Leveraging Feishu's native relationships, it inherently understands enterprise business data and can operate systems and tools according to existing standards and permissions by combining "skills" and "connectors." Most importantly, its data access follows Feishu users' identities and permissions, preventing boundary violations.

This native integration fundamentally differs from third-party integrations. Generic AI products lack understanding of a company's organizational structure, departmental divisions, or data statistics. Collaborative office platforms, however, retain communication records, documents, and business processes among employees—messy and highly fragmented but essential contexts for enterprise AI to understand operations.

This also constructs Feishu's competitive barrier against pure model companies: models can be replaced, and agent products iterate rapidly, but an enterprise's internal identity systems, permission settings, and work records are difficult to migrate in the short term (short-term). Once a collaboration platform becomes the primary channel for agents to access information and execute tasks, its client binding becomes stronger and deeper than traditional office software.

The launched "Doubao Work Partner" elevates Feishu and Doubao Work's native integration to a new level: expanding the service target from individuals to entire teams, upgrading agents from personal assistants to team members. Officially defined by Feishu as "China's first team intelligence product," it possesses independent identity, permissions, and memory, joining group chats and meetings like a colleague. Enterprises uniformly configure its permissions, capabilities, and usage limits, enabling All staff collaboration with the same agent without individual setups. Within authorized scopes, it can use Feishu documents, spreadsheets, meetings, calendars, and enterprise business systems to synchronize information, review documents, organize weekly reports, analyze code, and proactively advance projects by accumulating and reusing team skills from group chats, internal documents, and meeting records.

The on-site demo of Feishu's internal work partner "Xiao Fei" showcased specific capabilities: it proactively identifies project issues, analyzes and attempts fixes; follows up in group chats based on member replies; requests authorization from higher-permission members when encountering cross-hierarchy permissions and feeds results back to the group. Skills honed by one person can be reused by the entire team. It independently completes tasks like monitoring group issues, troubleshooting code, summarizing weekly reports, and processing customer feedback.

Currently, Doubao Work Partner is still in a Targeted co creation (directed co-creation) phase with enterprises and will gradually open to clients.

Breakthrough: Team Intelligence

Feishu disclosed key data at the conference: in the first half of 2026, Feishu's annual recurring revenue growth reached 2.5 times that of the same period last year, a new high since its inception. Over 90% of new clients simultaneously purchased Feishu AI products, and the total number of SME clients grew by 70% in the past year.

Significant progress was also made in industry penetration: currently, 9 out of China's top 10 listed automakers by market cap fully or partially use Feishu; all listed Chinese tea beverage groups use Feishu; 9 out of the top 10 humanoid robot sellers in the first half used Feishu. From automotive, consumer electronics, FMCG, and Pan Internet (pan-internet) to emerging industries like large models and embodied AI, Feishu has entered leading enterprises across sectors.

These data points confirm at least two core trends: first, AI is becoming a core decision factor—not just an added bonus—for enterprises purchasing office software; second, frontier tech industries embrace AI office solutions more readily, driving the evolution of office platforms in turn.

Notable shifts in client structure are also evident: while Feishu's early clients were mainly internet and tech companies, its clientele now spans traditional retail, consumer goods, manufacturing, energy, and biopharmaceuticals. However, varying digital maturity levels and diversified AI needs across these industries pose higher demands on Feishu's product adaptation and service capabilities.

From an industry landscape perspective, competition in the AI office sector has intensified. In six months, Tencent, Alibaba, and ByteDance completed AI office product line Integrate s (integrations): Tencent WorkBuddy rapidly gained traction after its March launch, directly driving documents, spreadsheets, emails, and calendars within WeCom via its open CLI interface; Alibaba consolidated QoderWork, Wukong, and MuleRun into "Qianwen Office" in August, prioritizing the enterprise market.

The three paths differ clearly: Tencent focuses on connectivity, Alibaba on organization, and ByteDance on collaboration. However, their commonality lies in shifting competition from functional comparisons of personal efficiency tools to deep organizational intelligence implementation. Simply equipping individuals with smarter dialog boxes is no longer the endpoint.

This explains why "team intelligence" became ByteDance's chosen breakthrough. Slack is essentially a connector, relying on third-party apps for documents, spreadsheets, and approvals, requiring cross-platform stitching of agent contexts. Feishu's advantage lies in its native integration of group chats, documents, meetings, calendars, tasks, multi-dimensional tables, and project management, with data, permissions, and workflows circulating within a single system. Here, agents need not piece together contexts across platforms or repeatedly adapt to different tool interfaces. Feishu's decade-long accumulation of "worksite" data has become a scarce asset in this round of competition.

Embarking Again in the Era of Intelligent Office

2026 marks Feishu's tenth year. From an internal tool to an industry-standard product, from human-to-human collaboration to human-agent collaboration, Feishu has completed a crucial identity transformation.

The signals from this conference are clear: ByteDance is escalating its AI bets from C-side traffic arenas to B-side productivity fields. Feishu is no longer just a collaborative office tool but is evolving into an enterprise work operating system in the AI era; Doubao Work Partner is not merely an AI assistant but is becoming a formal digital team member.

Challenges remain: team intelligence is a new category, and enterprise acceptance needs cultivation; agents' actual implementation effects must be validated in real business scenarios; issues like data security, permission management, and cost control will influence enterprise purchasing decisions.

But Xie Xin is confident. In his view, Feishu is the most beloved collaborative office platform, as well as the best working platform for Agents. Coupled with the enterprise-level Agent Doubao Work, which is natively integrated, it will be the best choice for work in the AI era. Just as he believed ten years ago: 'Give people exceptional tools, and they will create the extraordinary.'