Sixty-five percent of the code Anthropic's product team ships is now written by a tool the company only just handed to everyone else. That tool is Claude Tag, it launched on June 23, 2026, and it does not sit in a chat window on its own website. It lives inside Slack, where you summon it the way you would a coworker: type @Claude in a channel and hand it the work.
Anthropic is pitching this as a new way to work with Claude, not a new model. The same Opus 4.8 sits underneath; what changes is the posture. Instead of one person prompting a private assistant, a whole channel shares one Claude that everyone can see, steer, and pick up from. Full disclosure before we go further: we build a Slack teammate too, so weigh what follows with that bias in mind.
What is Claude Tag?
Claude Tag is Anthropic's AI teammate that joins a Slack channel as a shared participant. Anyone in the channel can tag @Claude to assign work, watch it run in the open, and continue where a colleague left off. There is one Claude per channel, not one per person. It is in beta for Claude Enterprise and Team customers.
It replaces the older "Claude in Slack" app rather than sitting beside it, and administrators get a 30-day window to opt in before the switch, per Anthropic's documentation. The headline is not the model. It is that Claude now has a seat in the room where the work already happens.
Persistent memory and ambient mode are the real shift
This is where Claude Tag stops behaving like a chatbot. Two features do the heavy lifting.
The first is memory. As Claude follows along in a channel it learns more about the work, and when an admin grants permission it can read other channels and connected data sources to gather facts on its own. A Claude living in a support channel slowly absorbs your products, your customers, and your escalation paths, so the next answer starts from context instead of a cold prompt.
The second is ambient mode. With it switched on, Claude posts without being asked: it flags things from across the organization, follows up on threads that have gone quiet, and keeps people updated on what it thinks they need to know. It can also schedule tasks for itself and pursue a project over hours or days, and you can DM it privately with your own connectors attached.
Picture the difference in practice. Today, an incident channel goes silent at 2am and someone reconstructs the timeline the next morning from scrollback. A teammate in ambient mode would have posted the running summary as the incident unfolded, pinged the on-call owner when a thread stalled, and had the postmortem draft waiting. None of that requires a smarter model. It requires a model that is already in the channel and allowed to act.
How admins keep a shared Claude from leaking context
A shared bot with memory that can read across channels is also a governance problem, and Anthropic clearly knows it. The answer is scoped identities. Administrators define which tools, information, and channels each Claude can touch, and its memories stay inside that scope. The example in the announcement is blunt: a Claude set up for sales work will not pass its memories to one set up for engineering.
There are spend controls to match. Token budgets can be capped at both the organization and the channel level, and every action is audit-logged, including who asked Claude to do it. That last detail matters more than it sounds. The moment a bot acts on its own in a shared channel, "who told it to do that" becomes a real question, and a logged answer is the difference between a teammate and a liability.
What Claude Tag means for teams already using AI in Slack
TechCrunch frames Claude Tag as Anthropic moving into the context layer - the same ground Microsoft Copilot, Glean, Snowflake, and Databricks are all fighting over. The prize is not a smarter model. It is being the system that quietly knows your company: its decisions, its history, where the doc lives.
For most teams the practical question is consolidation. If you already pay for Claude Enterprise or Team, a capable Slack teammate is now bundled in, and "good enough and already included" beats a separate tool more often than vendors like to admit. The counter-argument is lock-in: a teammate tied to one model vendor and one chat app is a narrower bet than one that works across Slack and Microsoft Teams and stays neutral on which model answers. That is the tradeoff worth arguing about internally, not which demo looked slicker.
A few caveats keep the launch honest. It is beta, Slack only for now, and limited to Enterprise and Team plans. And that 65 percent figure deserves a second look.
The bigger story is not that Anthropic built a Slack bot. It is that the major model labs now agree the interface for AI at work is not a separate destination you visit. It is a teammate that shows up in the channel, remembers what happened last week, and occasionally speaks first. The hard part was never generating text; it was earning a standing invitation to the channel and knowing when to use it. Whether that feels like help or like a coworker who never logs off is the thing every team is about to find out for itself.