WorkSwarm: The Ice Fishing of AI Teams—A New Paradigm for Office Agents

When AI Agents Become a Team

Office agents are the hot ticket right now. Everyone's stuffing smarter AI into word processors, spreadsheets, and meeting schedulers. But real work rarely ends after a single generation. Gathering info, drafting, revising, handing off files, syncing progress—the more complex the task, the more roles and handoffs you need. When AI starts handling these jobs, a new question pops up: Can multiple agents and humans collaborate like a team, passing the baton and building on each other's work?

openJiuwen recently took a swing at this with WorkSwarm, a swarm of office agents that brings multi-agent collaboration into personal office scenarios. Built by Huawei's 2012 Lab, Huawei Cloud, Devices, and Computing, openJiuwen announced last week that its swarm agents have been upgraded to WorkSwarm, first landing on the HarmonyOS PC app market with support for Windows and Mac.

To see what it can do, start with a few fun demos. Seven agents each handle lyrics, composition, arrangement, and vocals to produce a full song. Or two AI writers and one human share a single Word doc, taking turns to continue writing the classic "Yueyang Tower Record."

Behind the Scenes: A Coordination Playbook

What ties these demos together is the same agent collaboration logic. WorkSwarm assembles agents with different skills based on your goal, assigns tasks, and lets each member share the project background and latest outputs, then picks up where the last one left off. You can check progress, tweak requirements, or jump in yourself to complete a step.

For tasks of varying complexity, WorkSwarm offers two modes: single agent and swarm. Lightweight chores like queries or text tweaks go to one agent. Complex tasks that need multiple roles and rounds of deliverables are handled by a team of agents. All member discussions, task statuses, execution logs, and project files stay in one workspace.

So instead of a single AI assistant, you get a team of agents working alongside you.

From Solo Grind to Seamless Handoffs

Writing a proposal means researching, brainstorming, drafting, reviewing, and formatting. Making music requires someone for lyrics, melody, instrumentation, vocals, and a test listen. Multi-author writing brings version control, progress tracking, and handoffs. So much time gets lost in the gaps between tasks.

WorkSwarm tackles this with something it calls Coordination Engineering. In this setup, some agents plan, others execute, and a few specialize in spotting problems. Upstream outputs—docs, audio, logs—flow naturally to downstream members. You can sit in the "lead" seat or take on a specific task yourself.

The interface is straightforward: watch discussions in a group chat, track progress on a task board, find files in the project area. Want to know what a member did? Open their execution log. Spot a wrong turn? Add a quick note.

Office Work: A Team That Delivers

Imagine you're on a high-speed train and get an urgent briefing request. You pull out your phone, send a few messages in Feishu, and a reporting team spins up on your HarmonyOS PC. You're still on the train, but the work is already churning on your computer.

If the outline doesn't cut it, you can send it back. For technical details, @mention an expert agent for a consult. Midway, you can add a polish specialist. You're no longer a data courier—you're the decision-maker in the room.

The bigger pain point is batch material creation. In swarm mode, you input a topic, audience, and style, and the system auto-assembles a team that splits the work—research, structure, content, and integration run in parallel. What used to be a serial process becomes a parallel task chain, and you get a 200-page high-quality PPT in 20 minutes.

Code: Bring Your Ideas to Life

Swap to the Code space, and it's the same team play. Want to relive a classic arcade game like Tank Battle that's nowhere to be found? Swarm mode assembles a dev team: frontend and backend work on separate branches, merge to main, testers gate, a leader coordinates. Midway, if a new idea strikes, @mention a member to add a feature.

Or a calorie tracker? Just describe what you want in a sentence. The system plans the tasks, writes ArkTS code, compiles it, and produces an installable app for HarmonyOS PC.

Seven Agents Walk Into a Recording Studio

The latest demos give this workflow a tangible shape. The first video starts with a simple musical request. The user wants a coastal typhoon theme with cinematic, narrative vibes. WorkSwarm quickly pulls together a music team: a team lead, lyricist, composer, arranger, lead singer, accompanist, and interlude specialist—seven roles in total.

Each has a clear job. The lyricist writes lyrics and singing cues. The composer sets melody and structure. The arranger handles instrumentation. The lead singer focuses on articulation, breathing, and emotion. The interlude specialist checks transitions between sections. The accompanist suggests harmonies and dynamics. The team locks in a style—Epic Cinematic Dark Pop—and maps out Intro, Verse, Chorus, Bridge, and more. A first demo version is ready in no time.

The real magic happens after listening. The lead singer feels the emotion could push higher in spots. The interlude specialist goes section by section to check transitions. The accompanist offers a completeness judgment. They debate the Bridge's delivery, the Chorus's drive, the interlude's length, and the climactic beat, then feed the notes back into the creative loop. In the video, the version ticks from music-2.0 to music-2.6, and every change—what and why—is logged in the project.

Final deliverables include MP3, lyrics, composition plan, arrangement summary, singing instructions, accompaniment suggestions, interlude review, version history, delivery summary, and style prompt—12 files in all. If you want to tweak later or reuse the style for a new track, everything is ready to go.

The song in the video goes through the full cycle: create, listen, critique, revise. Each agent's professional lens lands on concrete files and versions, making team collaboration visible.

AI Writes, Human Writes, Phone Joins the Relay

The second video feels like a casual collab experiment. The team is just three: two AI writers and one human writer. The task: take turns continuing "Yueyang Tower Record," one sentence per round, signing each sentence with your name.

The three share a single Word doc on the desktop. When it's an AI's turn, it reads the existing content, finds where to pick up, writes a sentence, and marks it "AI Poet 1" or "AI Poet 2." Then it saves, closes, and reopens Word, and pings the next member in the group chat.

The human writer joins via an invite link on their phone. When it's their turn, they get the previous text and the current task. They type lines like "Embracing distant mountains, swallowing the Yangtze" and sign "Human Player," then hand the baton back. The desktop Word updates instantly, and the chat logs the handoff.

Closing and reopening Word might seem trivial, but it mirrors daily office work. The agent has to understand writing instructions, handle save/switch/notify. The human has an identity, tasks, and a message history. They can contribute or correct content anytime. "Human on the Swarm" and "Human in the Swarm" become natural actions: check progress, take a task, write, then call the next one.

Four Pillars Under the Hood

1. Autonomous Team Formation and Task Orchestration

WorkSwarm matches roles to goals, breaks down workflows, and organizes collaboration. For complex tasks, you don't have to summon tools one by one or manually ferry outputs.

2. Shared Context and Seamless Relay

Lyrics, arrangement plans, demo versions, and Word docs become shared team assets. Each member builds on upstream outputs, cutting redundant communication and info loss.

3. From Chat to Real Execution

Agents don't just generate text. They read and write files, operate apps, apply specialized skills, and deliver finished products. Work happens in real tools and real documents, not just in a chat window.

4. Transparent Process and Continuous Evolution

Member roles, task statuses, review comments, and version changes are all observable and traceable. Successful role combos and workflows can be saved as a Swarm Skill, turning one experience into reusable capability for the next task.

Your Next Colleague Might Be a Whole Team

WorkSwarm offers both work and development spaces, supporting HarmonyOS PC, Windows, and Mac. With a mobile entry point, you can stay in the loop when you're away from your desk—read messages, make calls, or jump into a task.

For individual users, it takes over data shuffling, process coordination, and file ops, freeing you for judgment and creation. For teams, a proven workflow can be saved, tweaked, and reused across projects. The longer you use it, the more it becomes a team of agents that knows your business, plus a work style that's truly yours.

One demo finishes a song. Another keeps a story going. Neither aims for spectacle, but they make WorkSwarm's value crystal clear: it can summon members, carry work forward, use tools, and always leaves a seat for you.

WorkSwarm—One Platform, Super Teams. Complex work, handled by a team that clicks.

openJiuwen recently launched an experience environment with a pile of free tokens, and it's recruiting developers and evangelists. Contributors can earn monthly community rewards, and outstanding devs can apply for official Huawei certification. The community has a growth ladder from AI enthusiast to senior evangelist, with chances to speak at events and shape the ecosystem.

If you're an AI tinkerer or an aspiring tech leader, openJiuwen welcomes you. Check out the experience at openjiuwen.com or dive into the code on GitHub.