Tiangong 3.2 launches Skywork Tags: AI Agent joins your team chat as a colleague

Kunlun Tech's Tiangong AI released a major upgrade to version 3.2 on July 2, introducing Skywork Tags — a feature that lets teams invite the Skywork AI agent directly into their group chats as a collaborative team member.

Unlike most existing AI assistant products that require users to export and migrate their team context into a separate workspace, Skywork Tags takes the opposite approach. Users simply add the Skywork bot to their existing Slack, Feishu/Lark, DingTalk, Discord, or Telegram group, and invoke it with an @mention.

The core design philosophy of Skywork Tags is shared rather than private. Only one Skywork exists per channel, facing everyone equally — not individual assistants for each user, but one shared team member. This creates three key changes: transparency (everyone in the channel sees what the AI is working on), handoff capability (one person starts a task and a colleague can seamlessly continue), and compounding improvement (the AI accumulates team context over time by staying continuously present).

According to a report from QbitAI, the Tiangong team ran an internal controlled experiment over several months. They deployed two types of Skywork bots — personalized individual versions and one shared instance in a 100-person team group. Initially, the personalized versions performed better. But after two to three weeks, the shared version overtook them, fed by dozens of daily conversations from diverse perspectives across the team.

The approach parallels Anthropic's recently launched Claude Tag feature. Andrej Karpathy characterized this pattern as the third paradigm shift in LLM interaction — from web to desktop app to AI as a persistent entity within an organization.

Skywork Tags is now available. Chinese users can access it at tiangong.cn, and international users at skywork.ai.