After the National Day holiday, a TRAE developer found himself needing to take an extra step to return to his familiar programming interface.

He had been accustomed to using the TRAE IDE. However, after an upgrade, the client defaulted to opening in Agent mode. To continue coding, he had to manually switch to IDE mode and then reselect his previous project. What frustrated him even more was the lack of a clear setting to change the default startup mode. Additionally, after closing the window, a persistent tray icon remained. He took to the TRAE official community to inquire if there could be an option for those who solely wished to use the IDE.

On October 9, TRAE Work and TRAE Code officially merged. The former primarily focused on handling documents, data, and office tasks, while the latter catered to software development needs. The merged TRAE retains both Agent and IDE modes, aiming to allow users to define requirements, proceed through development, testing, and delivery, all without the need to open two separate clients.

Less than two months prior, ByteDance had completed a round of integration for its AI office products. The TRAE and Kouzi teams were fully integrated into the Doubao system, with existing office capabilities concentrated in Doubao. Meanwhile, TRAE IDE and CLI continued to develop as a dedicated programming product line. ByteDance explained that this adjustment aimed to strengthen collaboration on product and technical resources. The following day, the unified AI office product, "Doubao Work," was officially launched.

Earlier, on July 30, ByteDance had integrated the Feishu product team with the Doubao product team, placing both under the leadership of Doubao's head, Zhao Qi. The Feishu sales, marketing, and customer service teams were integrated with Volcano Engine. These adjustments were made to consolidate models, office products, and enterprise customer resources that were previously scattered across different teams.

While ByteDance streamlined its organizational structure, TRAE expanded its product capabilities. Based on the August arrangements, there was reason to believe that general AI office functions would increasingly concentrate in Doubao Work, while TRAE would continue to serve developers. However, the new October version did not adhere to this clear delineation.

TRAE not only retained its capabilities for documents, reporting, and data processing but also prominently positioned the workspace for managing multiple Agents within the product. Doubao Work, on the other hand, can also create PPTs, generate web pages and applications, and has begun supporting complex task planning and multi-Agent collaboration. Consequently, there remains significant overlap in the capabilities of the two products.

This raises the question: If Doubao Work can already handle these tasks, why does TRAE need to duplicate them? And if TRAE's primary purpose is programming, why does it incorporate so many non-programming tasks?

TRAE was originally designed as an AI programming tool for developers.

In January 2025, the international version of TRAE was launched. Six months later, it began allowing AI to take over more programming steps. The SOLO mode, introduced on July 21, could understand project requirements, generate code, call tools, run programs, and check results.

Previously, developers wrote code while AI provided suggestions. In SOLO mode, users could directly assign a task, and AI would decide the order of operations, with humans returning later to review.

TRAE thus attracted a group of highly engaged developers. The 2025 annual report, released on December 26, showed that TRAE had over 6 million global registered users and over 1.6 million monthly active users. Around 6,000 users wrote code with TRAE for more than 200 days a year.

Problems arose when AI began handling entire development tasks, with coding becoming just one part of the process. In TRAE's October product introduction, an example was given of an event registration website. Developers needed to organize registration rules, decide on page content, handle user-submitted forms, write code, test functionality, and finally prepare documentation.

If the website was only for a conference demonstration, generating a functional page might suffice. But once in actual use, the work didn't end there. As registration numbers grew, new filtering conditions needed to be added; if users submitted incorrect information, forms had to be modified; if event rules changed, the original code had to be adjusted.

Previously, such tasks were scattered across different software. TRAE Work handled documents, spreadsheets, PPTs, and other office tasks, while TRAE Code was responsible for code editing and debugging. Transforming a project from requirements into software and then into a deliverable often required switching back and forth between two clients. Requirements were in one place, code in another, and test results might remain in a chat history.

For developers, the real hassle of switching tools was figuring out where they had left off in their previous work.

The October 9 merger aimed to address this issue first. The new TRAE combines the Agent workspace with IDE. Users can manage multiple projects in Agent mode, assigning different Agents to handle requirements, code, testing, and documentation. When they need to scrutinize the program, they switch to IDE to view files, modify code, and debug. Automated tasks, templates, and project outputs are integrated into a unified workflow. Code changes, HTML, Word, PPTs, and reports can all be viewed centrally in "My Outputs."

From this perspective, TRAE did not suddenly decide to enter the office market. The development work it originally served was becoming less easily separable from office tasks. However, these capabilities are not unique to TRAE. Doubao Work, released on August 25, can already handle documents, PPTs, data analysis, and lightweight application generation. Users can operate computers, call external services, and assign tasks to multiple Agents within authorized scopes.

On September 23, Doubao Work introduced "Goal Mode" and "Plan Mode." The former allows users to set acceptance criteria in advance, with Agents checking results before delivery. The latter generates a task plan first, which users review, modify, and confirm before execution. Both modes aim to reduce deviations during long-term Agent task execution.

If the tasks involve organizing requirements, generating web pages, and preparing presentation materials, both products already have the necessary capabilities.

New Position assigned 12 simulated employee feedback items to both Doubao Work and TRAE, requesting the creation of an offline management dashboard and delivery of review instructions and acceptance records. Both sides used the same materials and instructions, retaining their respective starting modes.

The client showed that Doubao took 3 minutes and 45 seconds in the first round, while TRAE took 19 minutes and 25 seconds. Statistics, searches, and filtering all passed independent checks. Subsequently, we added the same set of modifications: closing old feedback, adding new feedback, incorporating overdue filtering, and ensuring status changes persisted after refresh. Both sides completed the tasks, with Doubao taking 4 minutes and TRAE taking 18 minutes and 5 seconds.

TRAE conducted browser testing and fixed an overdue annotation omission; Doubao did not perform browser testing but passed independent checks of these features. For managers who only wanted to organize feedback and prepare reviews, Doubao Work delivered a usable, modifiable tool faster.

At least for this small tool with only 12 simulated feedback items, a professional development environment did not demonstrate irreplaceable advantages.

If organizing requirements, generating web pages, modifying functions, and preparing review materials can all be done faster with Doubao Work, at what point do developers need to open TRAE?

New Position's test consisted of only two rounds of tasks. While the feedback dashboard passed functional checks, it had only 12 simulated data points, was not connected to a real business system, and did not need to handle different employees' accounts and permissions. If truly handed over to a department, the subsequent challenges would likely be far more complex than generating a web page.

For ByteDance, retaining two entry points may not be unreasonable. While both products can accomplish similar tasks, they have different accumulated resources and address different user needs upon entry.

Doubao Work's advantage lies not just in its model capabilities. On July 30, when the Feishu product team integrated with Doubao, ByteDance explicitly stated that Feishu's existing products and services would remain unchanged. By then, Doubao Enterprise Edition was already in internal testing with some Feishu clients, supporting the use of Feishu documents, spreadsheets, meetings, and group chats within authorized scopes. On September 15, Feishu launched the Doubao Work Companion, attempting to give Agents independent identities, permissions, and memories to participate directly in team collaboration.

Liang Rubo also explained at the Feishu Future Infinity Conference & Doubao Work Kickoff Conference the reasons for integrating Doubao, Feishu, and Volcano Engine. He believed that once Agents reached a usable stage, models, Agents, and collaboration environments must be more tightly integrated.

This gave Doubao Work access to a resource TRAE lacked: the work relationships enterprises had already accumulated.

Take the feedback dashboard tested by New Position. If it were truly integrated into a company's work system, employee feedback might come from Feishu spreadsheets or be scattered across group chats and documents. Rules for which feedback was handled by whom and which content could only be viewed by managers would already be in place with corresponding account and permission settings.

If Agents could read these materials within employees' existing permissions, there would be no need to repeatedly upload files or explain business contexts. Once feedback was processed, it could remain in the original collaboration environment for relevant colleagues to view and modify.

Users might not need to rebuild an office environment from scratch. What they need is for AI to understand how the company originally worked and complete part of it for them. Feishu's years of accumulated accounts, member relationships, business materials, and permissions are exactly what Doubao Work can leverage.

However, being able to read company materials does not mean the dashboard can be used long-term within the company. After employees submit new feedback, can data update automatically? If an administrator closes a feedback item, will others' pages sync changes? If a business department proposes new statistical requirements, how should the original program be modified?

As these issues arise, the small tool originally intended for demonstration might evolve into a software project requiring continuous development. This is where TRAE's original development environment could demonstrate more value.

Developers can start from the Agent workspace and, as the project becomes more complex, switch to IDE to check code, modify files, and run tests. If one change affects other functions, they can return to the original project to locate the issue rather than regenerating a web page from scratch.

However, having a professional IDE does not guarantee better project maintenance. In New Position's test, TRAE conducted browser testing, while Doubao Work did not, but both ultimately passed functional checks. Two rounds of tasks were insufficient to determine which product would be more reliable for projects requiring continuous modification over several months.

The boundaries between the two products are blurring not only in development capabilities. According to TRAE's official documentation, after user authorization, Feishu CLI can directly read cloud documents, multi-dimensional tables, and other materials, as well as modify corresponding content. Developers can even read requirement documents from Feishu in TRAE, generate code and test cases, and write development progress back to Feishu.

Doubao Work is beginning to take on development tasks, while TRAE is entering enterprise office environments. The two products are converging from different directions.

However, having access to the same materials does not mean they can seamlessly continue each other's work. If an employee organizes requirements in Doubao Work, a developer subsequently modifies the program in TRAE, and it is finally handed over to the business department for acceptance in Feishu, this involves not just several files but also task records, code versions, and which content different employees can view and modify.

While TRAE can already operate Feishu materials within authorized scopes, publicly available information does not sufficiently prove that task statuses, conversation histories, and project files are fully interoperable between TRAE and Doubao Work.

Even within TRAE's own office assistant, such connections (handoffs) have limitations. Official documentation states that conversations initiated through different communication tools do not automatically merge contexts. In group chats, currently only the creator of the office assistant can delegate tasks; other members can only view records.

Doubao Work Companion attempts another approach: giving Agents independent identities, permissions, and memories to act as collaborative entities within the team. However, this product capability still needs to be tested in more real enterprise environments. Being able to issue instructions to TRAE from Feishu is not the same as TRAE truly taking over a team's task.

ByteDance has brought the two teams closer organizationally and enabled both products to access increasingly similar capabilities. However, for users, the problems do not automatically disappear. Requirements organized in Doubao Work today might need to be handed over to TRAE for development tomorrow; tools completed in TRAE must ultimately return to Feishu for business departments to use.

If switching products each time requires re-explaining tasks, confirming files, and handling permissions, ByteDance's two retained entry points will still force users to do extra work.

ByteDance can continue expanding each product's capabilities, but users' work will not automatically know where to start or end just because the teams have integrated.

On October 9, a developer voiced dissatisfaction with the latest iteration of TRAE, expressing a desire for simplicity: the ability to open the software and dive straight into coding.

Yet, during New Position's testing phase, Doubao Work demonstrated superior speed in crafting a feedback management dashboard, with its second round of modifications also successfully passing muster. For such straightforward tasks, at least, users currently see little incentive to opt for TRAE.

Another pertinent query arises regarding the retention of new users. Should an operator employ TRAE solely for a one-time webpage demonstration, the distinction between it and Doubao Work remains negligible. However, if that webpage evolves into an internal system necessitating ongoing enhancements, code refinements, and bug resolutions, TRAE's inherent development environment may gain greater significance.

TRAE faces the challenge of catering to both user segments: seasoned developers entrenched in IDE workflows and novices who might initially approach it from an office-based perspective before gradually requiring more sophisticated development tools.

Should Doubao Work progressively acquire more robust development, testing, and maintenance capabilities, the scope of work TRAE can sustain may dwindle further. Conversely, if users exhibit a willingness to continually refine and uphold an application within TRAE, rather than merely generating it and moving on, TRAE will preserve its standing as a distinct and valuable entry point.

*The featured image and accompanying illustrations within the text are sourced from the internet.