WorkSwarm: When AI Agents Actually Hand Off Work Like Humans

One AI Assistant Gets Lonely

Most office AI today works like a solo intern: you ask, it answers, you copy the result into a doc and start tweaking. But real projects aren't single-shot tasks. You gather research, draft, revise, hand off files, sync progress. The more complex the job, the more roles and handoffs you need. That's where WorkSwarm, from the openJiuwen community, takes a different approach.

Instead of one assistant doing everything, WorkSwarm assembles a team of specialized AI agents that pass work to each other—and to you—like colleagues stuck in a meeting room. It's built by a coalition that includes Huawei's 2012 Lab, Huawei Cloud, and device and computing groups. The upgraded swarm agents recently landed on HarmonyOS PC's app store, with Windows and macOS support on the way.

Seven Agents Walk Into a Recording Studio

To see how this works, look at the music demo. The user gives a simple prompt: a coastal typhoon theme, with cinematic, narrative feel. WorkSwarm spins up a seven-member team—a team lead, lyricist, composer, arranger, lead vocalist, accompanist, and interlude specialist. Each has a clear job. The lyricist writes lyrics and singing cues, the composer locks melody and form, the arranger handles instrumentation, the vocalist focuses on diction and breath, the interlude specialist checks transitions, and the accompanist suggests harmonies and dynamics.

They settle on a style—Epic Cinematic Dark Pop—and map out sections: Intro, Verse, Chorus, Bridge. The first demo appears quickly. But then the real collaboration kicks in. The vocalist feels some emotions could push harder. The interlude specialist reviews each transition. The accompanist weighs in on completeness. They discuss the Bridge, the Chorus's momentum, the interlude's length, and the build-up to the climax. All that feedback flows back into the creative loop.

The version history moves from music-2.0 to music-2.6, and every change is logged. The final deliverable includes twelve files: MP3, lyrics, composition plan, arrangement summary, singing instructions, accompaniment suggestions, interlude review, version history, delivery summary, and a style prompt. It's not just a one-off output—it's a reusable project you can continue tweaking or use as a template for new songs.

Humans and AI Share the Same Word Doc

The second demo is more playful but just as telling. A team of three—two AI writers and one human—takes turns writing lines for Yueyang Tower, a classic Chinese essay. Each member writes one sentence and signs their name. They all work on the same Word document on the desktop.

When it's an AI's turn, it reads the existing content, finds the next spot, writes its line, and adds “AI Poet 1” or “AI Poet 2.” Then it saves the file, closes Word, reopens it (as agreed), and posts a message in the group chat to nudge the next member. The human joins via a mobile invite. When it's their turn, they get a notification with the text so far and the task—they write “holding distant mountains, swallowing the Yangtze,” sign as “Human Player,” and pass the baton back. The desktop Word updates instantly, and the group chat keeps a full handoff log.

Those small, mundane actions—save, close, reopen, notify—are exactly what real office work looks like. The agents aren't just generating text in a chat window; they're operating on actual files and following a shared protocol. The human is not an observer but a full participant with identity, tasks, and a voice in the conversation.

Four Capabilities That Make It Work

Behind these demos are four key design principles that WorkSwarm uses to coordinate multi-agent work.

1. Autonomous Team Formation and Task Orchestration

WorkSwarm matches roles to the goal, breaks down the workflow, and assigns tasks. You don't have to summon each tool or manually carry outputs between steps.

2. Shared Context and Seamless Handoffs

Lyrics, arrangement plans, demo versions, Word docs—all become shared assets. Each agent builds on the previous one's work, cutting down on redundant communication and lost info.

3. From Chat to Real Execution

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

4. Transparent Process That Improves Over Time

You can see who did what, when, and why. Successful role combinations and workflows can be saved as “Swarm Skills,” turning one project's experience into reusable capability for the next.

Practical Scenarios: From PPTs to Code

The demos are fun, but the real value shows up in everyday grind work. Say you're on a high-speed train and get an urgent request to prepare a briefing. You open Feishu on your phone, type a few lines, and the agent team on your HarmonyOS PC starts assembling the material. By the time you arrive, it's ready. You can reject an outline, @-mention an expert agent for a technical deep-dive, or add a design specialist mid-way. You're not the courier of information—you're the decision-maker.

For bulk deliverables, the cluster mode shines. Feed in a topic, audience, and style, and the system assembles a team that works in parallel—research, structure, content, and integration happen simultaneously. What used to be a linear process becomes a parallel pipeline. The source claims 200 pages of high-quality PPT in 20 minutes.

Programming gets the same treatment. In Code space, you can recreate a classic arcade game like Tank Battle. The swarm sets up frontend and backend agents that develop on separate branches, merge into the mainline, and run tests under a lead's oversight. Mid-way, you can @-mention an agent to add a feature. Or describe a calorie tracker in one sentence, and the system plans the tasks, writes ArkTS code, compiles it, and produces an installable app for HarmonyOS PC.

Your Next Colleague Might Be a Whole Team

WorkSwarm runs on HarmonyOS PC, Windows, and macOS, with a mobile entry that lets you monitor progress, make decisions, or jump into a task while away from your desk. For individuals, it takes over the drudgery of file shuffling, coordination, and repetitive actions, freeing you for judgment and creativity. For teams, a proven workflow can be saved, adjusted, and reused across projects.

What's striking about the demos is how ordinary they are. Making a song, writing an essay—these aren't flashy AI stunts. They're everyday creative tasks that need iteration, feedback, and collaboration. WorkSwarm makes that visible: you see who did what, why a version changed, and what's next.

The openJiuwen community is now inviting developers and evangelists, offering free tokens and an experience environment at openjiuwen.com. There are incentives like the “Spark” and “Newcomer” awards, and top contributors can earn Huawei official developer certification. The community has a growth path from open-source enthusiast to senior evangelist, with opportunities to speak at events and shape the ecosystem.

If you're curious about where multi-agent AI is heading, this is a hands-on look. It's not about one super-intelligent assistant. It's about many specialized agents working together—and leaving a seat at the table for you.