ByteDance's Coze has officially released version 3.0, representing a major milestone in cross-platform AI collaboration. Spanning iOS/Android, Mac/Windows, and Web interfaces, Coze 3.0 moves away from the paradigm of a single all-powerful AI chatbot. Instead, it champions "project-centric, multi-agent collaboration," allowing users to assemble teams of specialized agents to tackle complex tasks collaboratively.
Multi-Agent Teamwork: Orchestration via Mentioning (@)
Under the new Coze 3.0 framework, users can create a unified project space and summon multiple specialized agents. In our hands-on test to develop an "AI Hotspot Tracking Dashboard," we assigned four agents: Coze’s built-in assistant "Xiaoyao," alongside custom-built Content, Product Design, and Frontend Development agents. By simply using "@" to mention these agents with a prompt, they communicated and delegated tasks automatically. The system broke down a complex requirement—from factual cross-referencing to full frontend React/Tailwind code generation—into a pipeline of continuous team labor, rather than outputting a generic, long-form chatbot response.
Beyond platform-native and cloud-custom agents, Coze 3.0 now supports one-click integration of local, developer-favorite tools such as Claude Code, Codex CLI, and OpenClaw, significantly enriching the diversity of the collaborative agent ecosystem.
Product-Level Generation and Multi-Modal Creation
To further evaluate its technical capabilities, we tested generating an interactive "Desktop Pet for AI Editorial Teams." Coze 3.0 smoothly translated abstract product concept ideas into neat React components with rich states (idle, thinking, alerting, and encouraging). Moreover, its multi-modal video capabilities have been upgraded. In a video generation test, we fed an generated article into the video project workspace, which, driven by the Seedance 2.0 model, automatically compiled a synced 45-second short video complete with assets, layouts, and subtitles.
Mobile-to-PC Remote Control: Bridging the Hardware Gap
The most striking highlight of this release is cross-terminal orchestration. We placed an NVIDIA financial report PDF on a physical computer desktop. Using the Coze mobile app on a phone, we instructed: "Analyze the financial report on my computer desktop." Granted permissions, the mobile agent successfully accessed and interpreted the local PC file. By seamlessly bridging desktop and mobile environments, Coze 3.0 transforms the mobile phone into a remote control for desktop-based agents, bypassing the hassle of manual file uploads and clipboard copying.
[AgentUpdate Depth Analysis] The launch of Coze 3.0 marks a significant pivot of AI Agents from simple chat interfaces toward a collaborative "Agent Operating System." While previous platforms focused purely on the raw reasoning power of individual LLMs, Coze 3.0 addresses real-world enterprise bottlenecks by establishing project-based multi-agent coordination and cross-terminal synchronization. By enabling integrations of external local tools like Claude Code, Coze effectively positions itself as an orchestrator of heterogeneous agent ecosystems. Crucially, the capability of controlling PC-based agents and retrieving local files via mobile devices breaks the long-standing physical isolation of AI from local operating systems. Moving forward, the true differentiator in the Agent race will shift from foundational model scales to the seamless orchestration of multi-device terminals, private local data, and cross-platform agent registries.
