Category: Technology | Published: 2026-06-09
For the past few years, AI at work has meant one thing to most people: a text box you open, type a question into, and read the answer from. Useful, certainly. But still fundamentally a tool you go to, rather than something that participates in the work alongside you.
That model is changing. Anthropic has launched Claude Tag, an AI assistant designed to live inside Slack as a genuine shared teammate rather than a separate application. It does not wait to be opened. It joins your channels, learns your work over time, takes on tasks independently, and interacts with everyone on the team, not just the person who last typed at it. It is a meaningfully different vision of what AI at work can look like, and it is worth understanding what it actually does.
What Claude Tag Is
Claude Tag launched in beta for Anthropic's Enterprise and Team customers in June 2026. It is built on Claude Opus 4.8, the company's most capable model at the time of launch, which Anthropic describes as a more effective collaborator with sharper judgement, greater honesty about its own progress, and the ability to work independently for longer than previous versions.
The basic interaction is simple. Within any Slack channel where Claude Tag has been enabled, any team member can type @Claude and assign it a task. Claude then works out how to complete the request, breaks it into sequential steps, draws on the tools and business systems it has been authorised to access, and posts the finished result back into the conversation for the whole channel to see.
That last part matters more than it might seem at first.
The Multiplayer Element
Most AI tools are single-player experiences. One person opens the tool, asks something, receives a response, and that exchange is invisible to everyone else. Claude Tag is deliberately different. It operates as a shared presence within each Slack channel, meaning the same Claude instance interacts with every member of that channel, not just whoever sent the last message.
When someone asks Claude to do something, the request and the output are visible to the whole team. A colleague can pick up the thread, ask a follow-up, redirect the task, or build on what Claude has produced. The interaction becomes part of the team's shared conversation rather than a private exchange between one person and a tool.
Anthropion describes this as making AI at work feel much more like collaborating with a teammate than using a conventional assistant, and that framing captures something real about what changes when the AI is embedded in the group's shared space rather than siloed in individual sessions.
How It Actually Works
When Claude Tag receives a task, it does not simply generate a text response. It decomposes the work into a sequence of steps, then executes those steps using whatever corporate databases, code repositories, connected applications, and business tools it has been given permission to access.
For technical teams, that might mean pulling from a codebase, running tests, and committing changes. For operational teams, it might mean querying a database, compiling a report, and posting a summary. For sales or account management teams, it might mean reviewing recent customer interactions and drafting a follow-up.
Anthrophen says that around 65 per cent of its own product team's code is now generated through its internal version of Claude Tag. That figure is striking not because it implies any particular percentage is right for other organisations, but because it illustrates that the company building the tool has genuinely integrated it into its own core workflows at scale.
The Asynchronous Part Changes Everything
One of the most practically significant features of Claude Tag is that it works asynchronously. You do not need to sit and wait for a response. You assign a task, carry on with other work, and Claude completes it in the background, whether that takes minutes, hours, or in some cases days.
The system can schedule tasks for itself, meaning it can break a larger project into phases and execute them sequentially without needing prompting at each stage. Anthropic describes Claude Tag as being able to pursue a project autonomously over hours or days, checking back in when there is something meaningful to report.
There is also what Anthropic calls ambient mode, where Claude proactively monitors the channels it has access to and flags things it believes are relevant: an unresolved issue that has gone quiet, a task that appears to have stalled, a question that was asked but never answered. Rather than waiting to be summoned, it keeps an eye on what is happening and nudges when it thinks something deserves attention.
This is genuinely new territory for AI at work. Most enterprise AI tools to date have been reactive by design. Ambient mode makes Claude Tag proactive, which changes the nature of what it is and what governance you need around it.
How It Learns Your Team's Work
Claude Tag builds context over time as it participates in channels. It learns the terminology your team uses, the projects that are in progress, the decisions that have been made, and the way your business operates. That accumulated context means users do not need to re-explain background information every time they assign a task. Claude already knows the project, because it has been in the conversation from the start.
Where administrators permit it, Claude can also draw on information from other authorised channels and connected systems, pulling together a broader picture of the organisation's work before responding.
Isolated Identities per Department
For most businesses, the prospect of an AI with broad access to multiple channels and systems raises immediate questions about data boundaries and information security. Anthropic has addressed this through scoped identities.
Administrators set up separate Claude instances for different business functions. The Claude configured for your sales team has access only to the channels, tools, and data relevant to sales. The Claude configured for engineering has a completely separate identity, separate memory, and access only to what engineering should see. Information from one does not flow into the other, and users in one department cannot access data or context from another through their Claude instance.
Private Slack channels are off by default. Claude only participates in channels where access has been explicitly granted. Administrators can also set monthly token spend limits at both the organisation and channel level, and full audit logs record every action Claude takes and which team member requested it.
What This Tells Us About AI at Work More Broadly
Claude Tag is a single product from a single company, but it is illustrative of a broader direction in enterprise AI. The generation of tools that simply answer questions when asked is giving way to agents that participate in ongoing work, remember context, take initiative, and operate independently for extended periods.
For businesses that are still in the early stages of exploring AI, this shift is worth understanding now rather than later. The question is no longer just whether AI can produce useful output when prompted. It is whether your organisation is ready to work alongside AI that operates continuously, has access to sensitive business information, and takes actions autonomously.
That requires thinking about governance, access controls, data boundaries, and the appropriate level of human oversight at each stage, before you deploy rather than after.
If you are thinking about how to introduce AI tools into your business in a way that is practical, secure, and genuinely useful rather than just impressive in a demonstration, our AI Consultancy page covers how we help businesses navigate exactly that.