Today, “Doubao Work” officially launched as an Agent product designed for productivity scenarios.
After extensive hands-on testing, our most straightforward conclusion is:
Its deep integration with Feishu offers one of the best answers yet to the question of how Agents can enter the workplace.
As mainstream workplace Agents become capable of handling files, calling tools, and generating web pages—and their basic capabilities increasingly converge—the real differentiator may be who understands a company’s actual work best:
What was discussed in project groups, what was decided in meetings, how far a record in a multidimensional table has progressed, and which data each person can view and modify.
This enterprise context is becoming the next battleground for workplace Agents.
Feishu has already accumulated precisely this kind of information. Once Doubao Work is connected to Feishu, it no longer needs to start from scratch every time it encounters a company. Instead, it can continue working along the discussions, data, and workflows the team already has in place.
So in this round of testing, we wanted to find out not only whether Doubao Work could get the job done, but also:
What changes when an Agent begins to understand a company’s context?
Doubao Work Can Handle Whatever You Want an Agent to Do
When I opened Doubao Work, my first reaction was:
Wait— is this still Doubao? Why has even the logo changed?!
Yes, Doubao Work comes with an entirely new logo and its own standalone app.
The interface, however, retains the classic three-column layout: task management on the left, the execution process in the middle, and output preview and editing on the right.
One notable feature is that it supports one-click sign-in with a Feishu account.
If you have used Codex, Claude Cowork, or other workplace Agents, you will find virtually no learning curve.
All you need to do is clearly state your goal. Doubao Work will handle the task breakdown and tool calls on its own.
For our first task, we asked it to read brand materials on the desktop and create three promotional images, a 15-second video, and an interactive web page.
I did not upload the files individually or specify any tools, leaving Doubao Work to figure everything out for itself.
In less than 10 minutes, it returned all three deliverables:
The results showed that the images, video, and web page were all generated around the same set of brand information, while maintaining a consistent visual style.
The web page can also be edited and published further. Simply mark the area you want to change, and it’s updated—a very convenient workflow.
For the second case, we raised the difficulty a notch:
We gave it only a procurement brief and asked Doubao Work to independently find three office computer configurations and compare procurement options for 100 devices.
It used a visual browser to search for suppliers, then opened product pages and dynamic configurators to verify the details.
In the end, Doubao Work recalculated the quotes based on consistent configuration and service standards, calculated the complete cost per device, and generated a comparison table and an interactive dashboard.
For bulk pricing, inventory, and delivery times that could not be confirmed through public web pages, it did not fabricate the missing information. Instead, it marked them as “Quote Required” or “Estimated.”
From operating a visual browser and verifying public information to analyzing tables and building interactive web pages, Doubao Work handled everything quite smoothly.
After two rounds of testing, the conclusion was already clear:
Doubao Work can handle the tasks that mainstream workplace Agents can handle.
In fact, some of the details came as pleasant surprises.
For example, when a page could not be opened in the visual browser, it proactively returned to the search results and continued verification through another entry point. It also worked more naturally with Chinese websites and content platforms.
When you are away from your computer, you can use your phone to assign and review tasks remotely.
Labor-intensive tasks such as data collection and batch processing can also be handed off to a cloud computer to continue running in the background.
But at this point, a new question emerged:
As handling files, operating browsers, and building web pages gradually become standard capabilities, where will the real gap between workplace Agents emerge?
Enterprise Context Is Becoming the Dividing Line for Agents
The convergence of Agent capabilities has almost become inevitable.
Models can be swapped, tools can be integrated, and Agent Harness engineering is rapidly becoming open source. If you gain a new capability today, your competitors can add it within days.
The enterprise Agent race has therefore entered its second half. The first half was about who could get more work done; the second half is about who understands the organization better.
After OpenClaw became popular some time ago, an interesting pattern emerged: people would install the Agent, then immediately rush to connect it to Feishu.
Feishu also quickly launched official plugins and a CLI, opening up capabilities such as messaging, documents, and multidimensional tables.
The reason everyone chose Feishu is not complicated.
A company’s conversations, documents, meetings, business data, and collaborative relationships are all accumulated there. For an Agent trying to understand enterprise context, Feishu has become nearly impossible to bypass.
But there is still a difference between “connecting to Feishu” and “being natively integrated with Feishu.”
OpenClaw and other CLI Agents typically require users to install plugins, create an app, and configure identity and permissions separately.
By contrast, Doubao Work supports direct sign-in with a Feishu enterprise account. It has a natural home-field advantage: from the very beginning, the Agent operates within the same organizational identity and permission system.
After testing it myself, I found the experience genuinely seamless.
Take our editorial team, for example. We discuss the latest developments in the embodied intelligence industry in group chats almost every day.
Someone might share breaking news from the robotics field; someone else might add background on a company; and before long, the conversation has moved on to technical approaches and industry controversies.
There is a lot of information, and much of it is useful.
But when it comes time to actually plan a story, a problem emerges:
Who suggested that idea? Which one did we ultimately decide to pursue? And where did that crucial piece of information disappear to?
The only option is usually to painstakingly scroll back through the chat history.
This time, we signed into Doubao Work with our Feishu enterprise account and asked it to directly read the editorial team’s recent group chats about the World Robot Conference and embodied intelligence.
Doubao Work quickly located the relevant group-chat records and cloud documents in Feishu and began organizing the leads they contained.
It accurately extracted several key points previously summarized in the group. It also divided the leads into “This Week’s Priorities,” “Ongoing Monitoring,” and “Watchlist,” while adding owners and items requiring verification.
Even the reasons for topics that had been passed over were clearly documented.
Doubao Work then synced the results to a multidimensional table, generating an “Embodied Intelligence Story Radar” that could be further filtered, supplemented, and collaboratively edited.
The most convenient part was that the entire process required neither re-uploading the chat records nor explaining what the editorial team had previously discussed. Doubao Work organized everything on its own.
And scrolling through group chats is clearly not the only tedious task it can eliminate.
Doubao Work can also organize the progress of cross-departmental projects from group chats, tasks, and approvals, assess whether they are likely to launch on schedule, and identify where the risks are stuck.
In sales and operations scenarios, it can read Feishu spreadsheets, analyze target gaps or content performance, generate visual reports, and send next-step recommendations directly back to the relevant work groups.
Summarizing, analyzing, creating tasks, and finding the right people to move things forward—all of it takes place within Feishu.
That is the most tangible value of “enterprise context”:
Agents no longer need to start a new job every day.
One More Thing
As basic capabilities gradually converge, the next Agent race will be about who better understands an enterprise’s context and workflows.
Claude has chosen Slack, while Doubao Work is integrated with Feishu. Leading Agents are moving in the same direction: into the real communication, data, and workflows of the enterprise.
Getting work done is only the starting point for Doubao Work. It has already begun answering a more difficult question:
When an Agent enters a company, how can it truly become part of the organization?
The endgame for enterprise Agents has yet to be determined, and this competition has only just entered its second half.
But judging by its current product form and implementation progress, Doubao Work and Feishu are already among the fastest-running answers on this path.