Currently, most AI coding tools operate in silos. If a developer uses Claude Code for one task and OpenAI’s Codex for another, there is no native way for these sessions to share information. This forces the user to manually copy and paste context, decisions, and requirements between different windows. Agents Can Communicate (ACC) addresses this by providing a local communication layer that allows these disparate clients to 'talk' to one another on the user's machine.
The image illustrates the core concept of ACC: allowing disparate AI coding clients to communicate and share information.
ACC functions by connecting supported coding clients—including Claude Code, Codex, Antigravity CLI, Grok, and Kimi Code—to a shared coordination workspace. Unlike some multi-agent frameworks, ACC does not use a 'lead' agent to manage the others; instead, coordination runs locally, and the agents themselves decide when to communicate. This architecture allows for several workflows: a developer can start parallel tasks in different clients and let the sessions agree on shared API shapes, or request a 'second opinion' from a different vendor's model. For example, a Codex session can be asked to review a change made by a Claude Code session, with the findings being sent back automatically.
One of the most significant features of ACC is the 'handoff' mechanism. Rather than moving a full, messy transcript between sessions—which can disrupt the flow of a new conversation—ACC allows a session to leave behind specific decisions, verified items, and next steps. This ensures the receiving agent has the necessary context to continue work without being overwhelmed by the previous session's entire history. However, this design choice introduces a trade-off: only information explicitly recorded by the session is passed over. If a developer does not record a specific detail, it will not be available to the next agent. Furthermore, while ACC provides the channel for communication, it does not guarantee that agents will notice every dependency; the coordination relies on the agents' ability to identify and request the information they need.",