Ten multi-agent open-source projects have simultaneously entered developers’ view, covering programming, research, customer service, operations, trading, and other scenarios. Our judgment is that competition in AI products is shifting from “how well it answers” to “whether it can work as a team,” but the projects’ concentrated emergence does not mean the technical path is mature.
What this is
Agent Org Chart structures engineering, marketing, customer service, and sales as an organization; Grok Factory appoints a lead coordinator who assigns research, programming, and other tasks to different agents; Foreman detects blockers and verifies deliverables; and BOTROSTER supports shared files, browser operation, and scheduled tasks.
Most of these projects are built on models such as Grok. So-called multi-agent systems let multiple AI assistants, each responsible for part of the work, collaborate on a task. They have moved beyond chat-based Q&A to writing code, publishing content, making phone calls, booking restaurants, and trading securities, although many remain experimental repositories maintained by individual developers.
Industry view
Two changes deserve attention. First, AI is beginning to gain team-management, memory, and acceptance mechanisms rather than merely invoking tools. Second, “human confirmation of critical steps” is becoming the default boundary for real-world execution: PhoneZero, for example, will not complete every appointment on its own.
The risks are equally clear. The simultaneous emergence of ten projects shows growing developer interest more than it proves enterprise-grade reliability. Our only source is a Juejin project list that lacks data on actual usage, incident rates, permission isolation, and return on investment. Multi-agent systems can also amplify error propagation, runaway costs, and unauthorized execution, especially once connected to trading and phone systems.
Impact on regular people
For enterprise IT: System selectors should pay attention to multi-agent division of labor and acceptance mechanisms, but experimental projects should not yet be allowed to take over trading, customer service, or core data. Permissions and auditing must come before deployment.
For individual professionals: Research, writing, coding, and content operations will increasingly be completed through AI collaboration. Human value will shift toward defining objectives, validating results, and handling exceptions.
For consumer markets: More services will be able to book restaurants, make phone calls, and execute tasks continuously, but users still need to examine service terms covering pricing, liability for mistakes, and how personal information is used by multiple agents.