Most Slack bots have the memory of a goldfish. You tag them, they answer, and then they forget you exist entirely. Every new conversation starts from zero, which means you spend half your time re-explaining context that should already be obvious.
Anthropic is trying to fix that. The company has launched Claude Tag, a research preview now available to Enterprise and Team Slack customers. Instead of waiting to be summoned, Claude Tag runs as a persistent agent inside your channels, following conversations over time and building up an understanding of your team’s work as it happens.
The pitch is simple: an AI that reads every message while you were out and actually retained all of it. That sounds useful. It also raises some real questions that companies will need to think through carefully before deploying it broadly.
What Claude Tag actually does differently
The core mechanic here is persistent memory at the channel level. Traditional chatbots cold-start every session. Claude Tag does the opposite. It sits in your channels, accumulates context from decisions, threads, and updates, and builds a working picture of what your team is doing without needing a briefing each time.
Picture a mid-sprint status update. Someone tags @Claude and asks for a summary pulling from three different channels. Claude Tag already knows what was decided last Tuesday, which thread went quiet on Thursday, and what dependency another team flagged on Friday. No re-briefing required.
The specific capabilities the agent offers include:
- Persistent channel memory that learns team context over time automatically
- An ambient mode that proactively surfaces forgotten threads or stalled tasks without needing an @mention
- Cross-channel fact retrieval when granted permission through MCP, which stands for Model Context Protocol, a standard that connects Claude to external tools and data sources
- Scoped identities that let admins create role-specific agents for legal, engineering, or product teams, with no context sharing across those domains
- Multi-step task handling where the agent breaks complex requests into stages and posts progress back into the thread
Anthropic describes the model as an agent with “shared context, persistent awareness, and proactive execution” that works as a genuine member of the team rather than a tool you have to constantly manage.
Why this matters in a crowded market
Anthropic is not alone in chasing this idea. Microsoft Copilot, Glean, and Google Workspace AI are all working toward the same goal: becoming the persistent intelligence layer sitting inside enterprise software. The race is real, and the stakes are high for whichever company gets teams to rely on their agent first.
Claude Tag’s angle is Slack-native depth. Rather than adding a sidebar assistant that employees have to consciously switch to, it lives where the work already happens and participates without friction. That is a genuine differentiator compared to bolt-on approaches, particularly for organizations that run most of their coordination through Slack already.
The broader industry shift here is worth noting. Teams are moving away from AI as a search tool you query occasionally, toward AI as a participant that stays embedded in workflows continuously. Claude Tag is an early, concrete example of what that looks like in practice.
The governance questions companies cannot ignore
An agent that reads and remembers messages across your Slack channels indefinitely is not a simple tool rollout. It is a policy decision, and a significant one.
Slack admins do control access scopes and channel permissions, so there are levers to manage what Claude Tag can see. But that technical control only goes so far without clear organizational policies to back it up. Companies deploying this will need to answer some specific questions before they do:
- Which channels is the agent allowed to read, and who decides that?
- How long does it retain what it learns, and can that memory be deleted on request?
- What happens when an employee leaves, and their conversations are still part of the agent’s context?
- How does this interact with existing data retention and compliance requirements?
Enterprise security and compliance teams will push hard on these points, and rightly so. Employee privacy is a real concern when an AI is continuously observing team communication rather than being called in for specific tasks.
The technology is clearly moving faster than most governance frameworks. Companies that want to use Claude Tag responsibly need to build those policies now, not after the tool has been running in their channels for six months. Waiting for the next quarterly review is not a realistic option when the agent is already reading the room.
