Claude Tag: Anthropic gives Claude persistent memory in Slack
🔎 AI leaves the one-shot chat to become a full-time colleague
On June 23, 2026, Anthropic announced Claude Tag, a beta that redefines the place of AI in the daily workflow. No more need to open a tab, copy context, paste a prompt, wait for a response, and then report it back to Slack. Claude Tag lives inside Slack, continuously learns from authorized channel conversations, and acts as a persistent team member.
This isn't just another Slack bot. It's a paradigm shift: moving from AI as a sporadic tool to AI as a living organizational memory. Anthropic actually designed it as the direct evolution of Claude Code, its development agent, into a more proactive model adapted to teamwork.
The timing is not insignificant. Anthropic just signed an agreement with SpaceX for the Colossus 1 supercomputer — 220,000 GPUs and 300 MW dedicated to Claude. The infrastructure matches the ambition: making Claude an omnipresent agent, not just a simple chatbot. And for marketing teams and SMBs that live in Slack, the implications are concrete.
The essentials
- Claude Tag is a persistent AI agent integrated directly into Slack, available in beta for Claude Team and Enterprise plans.
- It features shared organizational memory: it continuously learns from authorized channels without needing to be provided with context manually.
- It performs multi-step tasks, proactively follows up on unfinished work, and gathers cross-channel information according to granular admin permissions.
- Each Claude instance is confined to specific organizational zones — no data leaks between departments.
- Claude Tag runs on Claude Opus 4.8, and Anthropic claims that AI now writes 65% of its internal code.
Recommended tools
| Tool | Main usage | Availability (June 2026) | Ideal for |
|---|---|---|---|
| Claude Tag | Persistent AI agent in Slack | Beta, Team and Enterprise plans | Slack teams with shared memory needs |
| Claude Opus 4.7 (Adaptive) | Versatile agentic LLM | API and Claude Pro/Enterprise | Complex reasoning tasks outside Slack |
| Claude Sonnet 4.6 | Fast and reliable general LLM | API and Claude Pro | Data analysis, writing, summarization |
| Claude Code Agent View | Development agent with visual dashboard | Available | Developers working in terminal |
What Claude Tag actually does — and what it doesn't do
Claude Tag replaces the old Claude in Slack app. The difference is fundamental: the old version responded when prompted. Claude Tag acts even when no one tags it.
In practice, a user can tag @Claude in a conversation to ask for an insight or assign a task. But Claude can also proactively join a conversation in ambient mode to flag a blocker, summarize a long exchange, or recall a decision made three weeks earlier in another authorized channel. All of this without ever having to be reprovisioned with context.
What doesn't change: the guardrails. Administrators define precisely which channels Claude can see. Each instance remains confined to its organizational perimeter. Marketing doesn't see what's happening in R&D, unless the admin explicitly allows it. Companies have 30 days to opt-in after activation, according to Anthropic's release notes.
The memory is not a raw dump of messages. Claude Tag builds a structured understanding of projects, decisions, and dependencies — what Rohit AI analyzes as a transformation of Slack into "shared AI memory" for teams.
Why Persistent Memory is the Next Big AI Shift
For two years, the dominant pattern of enterprise AI use has been one-shot chat. Open ChatGPT or Claude, paste in context, ask a question, get the answer. This model has two major problems: it is exhausting at scale, and it is amnesic.
The artificial amnesia of LLMs is not an insurmountable technical limitation — it was an initial design choice. Providers wanted to prevent the model from accumulating biases or sensitive data from session to session. But this choice created massive friction: users spend more time contextualizing the AI than exploiting its answers.
Claude Tag is part of a broader movement. Google introduced persistent memory in Gemini Live, allowing the assistant to remember a user's preferences and past conversations. The difference: Gemini Live's memory is individual. Claude Tag's is organizational.
This is a qualitative leap. Individual memory improves the personal experience. Organizational memory improves the coordination of an entire team. When Claude remembers that a marketing campaign was pushed back because the design was missing, and that this info was shared in #marketing-ops two weeks ago, it eliminates the manual information handoff that costs SMBs dearly.
The model under the hood: Opus 4.8 and Anthropic's strategy
Claude Tag runs on Claude Opus 4.8, a model that does not appear in standard public leaderboards but seems to be a variant optimized for long agentic tasks and extended memory. In the June 2025 agentic leaderboard, Claude Opus 4.7 (Adaptive) reached 94.3 points — third behind GPT-5.5 (98.2) and Gemini 3 Pro Deep Think (95.4). Opus 4.8 inherits this base with specific optimizations for persistent context.
Anthropic claims a remarkable figure: the AI writes 65% of the company's internal code. This is not anecdotal. It means that Anthropic is using its own agents in its own development on a massive scale — and that Claude Tag is the product of this intensive dogfooding.
This is consistent with their recent infrastructure strategy. The deal with SpaceX for Colossus 1 and its 220,000 GPUs is not solely aimed at training more powerful models. It also serves to support persistent agents that consume compute continuously — not just at the time of a prompt.
Persistent memory is expensive in terms of infrastructure. An agent that constantly "listens" to dozens of Slack channels and maintains a structured understanding does not operate on the same cost model as a one-off chat. Anthropic needs this computing power to make Claude Tag profitable at the enterprise scale.
Concrete implications for marketing teams
Marketing teams are among the heaviest Slack users in SMBs. Creative briefs, client feedback, exchanges with sales, content planning — everything goes through the channels. The problem: information lives, dies, and is buried in the history within a few hours.
Claude Tag changes this dynamic in three ways.
First, the automatic synthesis of briefs. A creative brief scattered across 47 messages in #campaign-q3 can be gathered into a structured document by Claude without anyone asking it to. It detects the "brief in progress" pattern and proposes a summary.
Second, proactive dependency tracking. If the landing page is blocked because the design assets haven't been delivered — info mentioned in #design-team — Claude Tag can flag it in #marketing-ops without human intervention. This is the ambient mode described by Dealroom : un agent qui rejoint proactivement les conversations.
Third, accelerated onboarding. A new marketer joining the team can ask Claude Tag about campaign history, positioning decisions, past failures — everything that has been discussed in the channels they have access to. Claude becomes the living wiki that no one ever has the time to maintain.
For SMBs that don't have a chief of staff or a dedicated project manager, this is a coordination gain that is disproportionate to the cost of a Claude Team plan.
SMBs and Slack: why Claude Tag could be the killer app
Large companies have sophisticated knowledge management tools — Confluence, Notion, Guru. SMBs have Slack. Not by strategic choice, but by default. Slack becomes the wiki, the makeshift CRM, the project manager, and the communication channel.
Claude Tag turns this default into an advantage. Instead of asking a 25-person SMB to maintain up-to-date documentation on top of working in Slack, Claude Tag makes documentation implicit. The information is already there, in the conversations. All it takes is an agent to structure it.
Windows Forum describes Claude Tag as an agent with "shared organizational memory, proactive ambient mode, cross-channel information gathering, and multi-step tasks". For an SMB, this translates to a tool that costs a fraction of a junior project manager but never sleeps, never forgets, and never goes on vacation.
The pricing model remains that of the existing Claude Team and Enterprise plans. No additional cost has been announced for Claude Tag in beta. This is strategic: Anthropic wants massive adoption before monetizing the added value.
Trust and governance: the real challenge of shared memory
An AI that remembers everything said in a company is exactly what scares DPOs and security managers. Anthropic understood this, and governance is the most developed pillar of Claude Tag.
Organizational containment is strict. Each instance of Claude Tag is limited to specific channels. The admin chooses the scope, and Claude cannot "drift" to unauthorized channels, even if a user asks it to in a conversation.
The 30-day opt-in period given to administrators is not a cosmetic grace period. It is the time needed to map sensitive channels, define scopes by department, and communicate to teams what Claude can and cannot see.
Rohit AI points out that teams must monitor several risks: the involuntary memorization of personal information, the possible surfacing of old data out of context, and the organizational dependency on an agent that could make mistakes. These are real risks, not theoretical ones.
Anthropic has not published a technical whitepaper on the encryption of Claude Tag's memory or on the exact retention of learned data. This is a point of friction for GDPR-bound companies that will want guarantees before deploying at scale.
Claude Tag vs Gemini Live : two visions of memory
Google introduced persistent memory in Gemini Live as an individual-centric feature. Gemini remembers your preferences, your past conversations, your personal context. It's an ego-centric memory.
Claude Tag is a socio-centric memory. It doesn't belong to an individual but to a team, a department, an organization. Claude doesn't remember "your" preferences — it remembers what the team decided, produced, reported.
The two approaches are not in conflict. They address different use cases. Gemini Live excels in the personal assistant — emails, calendar, personal search. Claude Tag excels in the coordination assistant — projects, decisions, team dependencies.
But for businesses, the socio-centric memory of Claude Tag is potentially more transformative. An employee who leaves the company takes their Gemini memory with them. The Claude Tag memory remains within the organization.
The ecosystem expands: beyond Slack
Anthropic is laying the groundwork. Reuters reports that Anthropic plans to extend Claude Tag to other platforms in the future, and EdTech Innovation Hub confirms this expansion intention.
Slack is the logical entry point. That's where corporate unstructured information is densest. But organizational memory isn't meant to remain trapped in a single tool. The probable roadmap: Microsoft Teams, then perhaps Google Workspace, then deeper integrations via API.
Claude Tag follows in the footsteps of Anthropic's work on persistent agent memory — as illustrated by Hermes' memory system, which shows how AI agents can maintain a structured state across interactions. Claude Tag is the enterprise application of this fundamental principle.
For technical teams, the convergence between Claude Code (development) and Claude Tag (collaboration) is notable. Anthropic is building an ecosystem of agents that share a common philosophy: context persistence. The terminal is no longer the only place where Claude acts — it acts wherever work happens.
Competitive positioning: Anthropic gains the upper hand in the enterprise space
OpenAI dominates agentic benchmarks with GPT-5.5 (98.2 points) and GPT-5.4 Pro (91.8). Google competes with Gemini 3 Pro Deep Think (95.4 agentic, 90 general). But when it comes to integration into real workflows, Anthropic has just taken a concrete lead.
Claude Tag is not a more powerful model — it's a more integrated product. And in the enterprise AI war, integration beats benchmarks. A marketing team doesn't need the model that scores highest on an abstract reasoning test. It needs an agent that knows its projects, its constraints, its deadlines.
TechCrunch nicely summarizes the positioning: "Anthropic's Claude Tag is learning your company, one Slack message at a time". It is precisely this continuous organizational learning capability that sets Claude Tag apart from a simple Slack wrapper around an LLM API.
❌ Common mistakes
Mistake 1: Confusing Claude Tag with the old Claude in Slack
The old app responded when queried, without memory between sessions. Claude Tag learns continuously, acts proactively, and maintains structured memory. It is not an update — it is a different product.
Mistake 2: Deploying without mapping sensitive channels
Claude Tag only sees what the admin allows. But if the admin allows #general by default without thinking, Claude will have access to HR, financial, or strategic discussions. Use the 30-day opt-in period to conduct a thorough audit of the perimeters.
Mistake 3: Expecting perfect memory from day 1
Claude Tag learns continuously, which means its memory is built over time. During the first few weeks, it will not have the necessary perspective on past decisions. It is an investment: the more the team uses it, the more useful it becomes.
Mistake 4: Ignoring GDPR risks
Even with organizational confinement, an AI that memorizes company conversations must be treated as a data processor. Document the flows, check Anthropic's terms, and inform the teams. Memory is an asset — but it is also a legal responsibility.
❓ Frequently Asked Questions
Is Claude Tag available for individual Claude users?
No. Claude Tag is in beta exclusively for Claude Team and Enterprise plans via Slack. Anthropic has not announced a timeline for Pro or Free plans.
Does Claude Tag completely replace the old Claude in Slack app?
Yes. Claude Tag replaces the old integration. The core features (answering questions, text analysis) are retained, but enhanced by persistent memory and proactive mode.
Which Claude model powers Claude Tag?
According to SaasRise, Claude Tag runs on Claude Opus 4.8, a variant optimized for long agentic tasks and persistent context. This model is not available separately on the public API.
Are the data learned by Claude Tag used to train public models?
Anthropic has not explicitly confirmed this for Claude Tag. Enterprise plans contractually include a data non-use clause for training. For the Team plan, check the current terms in the Claude Help Center.
Will Claude Tag work on Microsoft Teams or other platforms?
Anthropic told Reuters and EdTech Innovation Hub that it plans to extend Claude Tag to other platforms. No timeline was announced in June 2026. Slack is the launch point.
✅ Conclusion
Claude Tag marks the shift from AI assistant to AI colleague — an agent that remembers, anticipates, and acts within teams' actual workflow. Organizational persistent memory is not just an additional feature: it is the next standard for enterprise AI. For SMBs and marketing teams that live in Slack, the coordination gain is immediate. The remaining challenge is to master governance before memory becomes a risk. To follow the evolution of persistent AI agents, check out our guide des meilleurs outils IA updated every quarter.