Kunlun Wanwei Skywork 3.2 Releases Skywork Tags for AI Agents in Group Chats

Kunlun Wanwei Skywork 3.2 ships Skywork Tags that let AI agents enter live chat threads in Slack, Feishu, DingTalk, Discord and Telegram. Teams keep their current groups and simply mention the agent. Data stays in place. (Kunlun Wanwei announcement)

The update removes the need to copy messages into separate windows or new channels. Agents now read conversation history, files and ongoing replies inside the same space.

Skywork Tags therefore changes the integration point from separate apps to the chat itself. Users no longer migrate workflows.

Shared agents that absorb input from multiple people show stronger results than single-user agents tuned over weeks. One anonymized design team testing the feature reported that their shared agent surfaced past decisions from several members during a product review thread, cutting restatement time by more than half; they subsequently switched fully to the shared version. An anonymized engineering team at a mid-sized software firm found the shared agent automatically referenced prior API specification decisions made by different members during bug triage threads in Slack, reducing duplicate discussion volume by roughly 40 percent and shortening average triage time by 25 percent. A marketing team reported that the agent tracked cross-thread campaign metric updates inside a DingTalk group, allowing them to compile weekly performance summaries without manual collation and cutting report preparation time by approximately 30 percent; both teams later adopted the shared version exclusively. (Kunlun Wanwei announcement) Teams report they eventually switch entirely to the shared version.

Shared context grows naturally as the agent stays inside the group. It records decisions, links and follow-ups without extra effort from users.

The approach stands in contrast to tools that ask teams to start fresh in dedicated AI workspaces. Kunlun Wanwei keeps the original group intact.

Competitors such as Microsoft Copilot and Slack AI still route most work through bolted-on side panels or separate bots. Those panels require users to switch views or repeat context.

Skywork Tags avoids that switch. The agent appears as one more named member in the member list.

Several early groups tested the feature for two weeks. They tracked reply quality, task completion speed and the number of times people needed to restate prior facts.

Shared agents reduced restatement counts by roughly half. Completion time for routine follow-ups dropped in the same tests.

Critics note that shared context creates new data exposure points. A single misconfigured channel permission could surface internal notes to an agent that later joins a wider group.

Kunlun Wanwei states the agent only sees content inside the groups it is added to. Access logs are available to administrators.

The company also says no training runs are performed on customer chat data. Only inference occurs inside the active thread.

These claims remain unverified by independent audits. Enterprise security teams will watch for third-party reviews in the coming months.

The next signals to track are adoption inside publicly listed companies, any published retention or accuracy metrics, and moves by Slack or Microsoft to match the same in-thread presence model.