Anthropic has launched a new collaborative AI agent called Claude Tag, now in public beta on Slack. It lets enterprise and team subscribers treat AI as a persistent, asynchronous collaborator inside their workflows. Available for Claude Enterprise and Claude Team plans, it moves beyond one-on-one chat to allow multiple people to jointly direct a single AI within a shared channel.
Unlike traditional chatbots that only respond to direct prompts, Claude Tag embeds directly into Slack channels, capable of autonomous planning and async execution. When team members mention @Claude and submit a task, the model breaks it into stages, taps into connected code repositories, databases, or external tools, and posts results back in a thread when done. It also offers a context-aware mode that continuously tracks channel discussions, proactively offering information or following up on stalled threads.
On the security side, channel admins can set per-channel tool access and data permissions, keep model memories isolated, and enforce global or per-channel token limits. Internally, Anthropic has been using Claude Tag heavily—the product team says 65% of its code is now generated by an internal version.
Andrej Karpathy called it the third paradigm of large language model user interfaces: AI evolving from early web chat and local desktop tools into persistent, asynchronous collaborative entities with organization-level context and tools.