Anthropic reimagines Claude in Slack as nosy, always-on agentic AI coworker
The Register ● Covered by 3 sources
Anthropic swapped its old Claude Slack bot for 'Claude Tag,' an always-on AI that lives in your channels and remembers everything. It learns from your team's chats, codebases and tools indefinitely, and can even assign itself tasks to run for days.
Based on reporting by The Register — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Anthropic just retired its bare-bones Claude Slack connector and replaced it with something far more ambitious: Claude Tag, a persistent AI teammate that hangs around your channels indefinitely, soaking up context from whatever conversations, code, and data admins let it see. Tag @Claude and it delegates work like a human coworker would — except this coworker never forgets what you told it three weeks ago and never clocks out.
The pitch from Anthropic is collaboration, not convenience. Unlike the old connector, which basically gave you one-off replies, Tag is shared across an entire channel. Everyone talking to it is talking to the same instance, building a shared memory that persists and even bleeds across channels if given permission. Anthropic says its own product team now has 65 percent of its code written by Tag, and staff are spending more time assigning tasks to "many Claudes in parallel" than doing the work themselves — a framing the company clearly wants read as progress rather than a quiet admission that engineers have become middle managers for bots.
The more eyebrow-raising feature is ambient autonomy. Toggle it on, and Tag will proactively surface things it thinks you need to know, scheduling its own tasks to run over hours or days without a human prompting it each time. That's a meaningful step up from a chatbot that answers when asked — this is software that decides, on its own initiative, what deserves your attention next.
Anthropic is sweetening the transition with real money: $25,000 in credits for Enterprise Slack customers who migrate, $2,500 for Teams customers, both expiring September 1st, with the old connector shutting down entirely on August 3rd. That urgency lands awkwardly given Anthropic recently ditched bundled pricing for metered billing that's already been surprising customers with bigger-than-expected invoices — so nobody quite knows how far those credits will actually stretch once the free trial period ends.
The part that deserves more scrutiny than a press release will give it is data exposure. Letting an AI model permanently ingest your organization's Slack history, codebases, and support tickets for "contextual learning" means handing a lot of sensitive material to servers you don't control. Anthropic's own framing treats this as a feature. Whether your legal and security teams see it the same way is a separate question entirely.
My take — AI-written commentary, not fact-checked reporting
An AI that remembers everything you've ever said in a channel and decides on its own what you need to know isn't a productivity tool, it's a permanent surveillance layer wearing a helpful-coworker costume, and companies rushing for that $25,000 credit before September 1st are optimizing for a discount, not for whether they actually want Anthropic's servers holding a persistent map of their codebase and support tickets. This is the same pattern every closed-model vendor runs: dangle a shrinking window of free credits, get the data pipeline installed before anyone reads the fine print on retention. I'd rather see this kind of ambient agent running on something self-hosted, even if it's clunkier, than trust a subscription model that can change its pricing terms whenever it likes.
Read more about this at: The Register