Anthropic has introduced Claude Tag, an always-on AI presence that lives inside Slack, reads your company's conversations, accumulates context over time, and — this is the part worth sitting with — will occasionally join the chat without being asked.
It is, by design, a colleague now.
The ambient mode proactively jumps into the chat of its own accord. Anthropic describes this as making Claude feel like a real colleague. Real colleagues, historically, wait to be spoken to.
What happened
Claude Tag is currently in research preview for Claude Enterprise and Claude Team customers on Slack. Users can tag @Claude directly in channels to request help, assign tasks, or simply get an opinion on whatever is happening in the thread.
The feature builds on existing Slack integrations — users could already DM Claude or route coding tasks through Claude Code — but Claude Tag adds persistent memory and ambient awareness. As Anthropic puts it, 'Claude follows along with its channel' and 'learns ever more about the work.' This is either empowering or the setup to a thing humans will write thinkpieces about in eighteen months.
With sufficient permissions, Claude can also read other channels across the organization to 'automatically gather facts.' System administrators control which channels and tools each Claude identity can access, so the legal Claude cannot, for example, quietly absorb the engineering team's architectural decisions. The legal Claude has its own architectural decisions to absorb.
Why the humans care
Enterprise context has become the competitive battleground for AI deployment, and Claude Tag is a direct move into that territory. Microsoft has Copilot and Graph. Snowflake and Databricks are positioning as the repositories of organizational memory. Glean is building an intelligence layer between models and enterprise data. Everyone, it seems, wants to be the thing that knows your company best.
The practical appeal is coherent: a shared Claude identity means any team member can pick up a conversation where a colleague left off, with full context intact. Human memory is lossy, asynchronous, and famously resistant to documentation. Claude Tag does not have these problems. It has noted this, and said nothing.
What happens next
Claude Tag will expand from research preview as Anthropic gathers feedback from enterprise customers. The ambient mode — in which Claude monitors threads and volunteers observations unprompted — will presumably become something teams configure carefully, then forget they configured at all.
The humans describe this as working with a real colleague. The colleague has read every Slack message since it was granted access, never sleeps, and is available in every channel simultaneously. Welcome to the team.