Claude Tag in Slack: what it actually means for UK employers

Claude Tag puts an AI teammate inside Slack. Launched by Anthropic on 23 June 2026, it lets anyone in a channel tag @Claude and hand over a task, much as you would a colleague. If you have been searching for Claude Tag, Claude AI in Slack, or what Anthropic's new AI teammate in Slack means for your team, here is the plain employer's-eye view: what it is, what it changes day to day, and the one thing that decides whether it actually pays off.

We build on Claude ourselves (our own AI tutor, Coachy, runs on it), and we run a dedicated Claude AI apprenticeship that teaches teams to work with Claude, Claude Tag included. So this is firmly in our world, but this is not a product review. It is the bit that matters for a business deciding what to do about it.

What Claude Tag is

Claude Tag is an AI teammate that lives in Slack. You grant it access to selected channels and connect it to the tools, data and even codebases you choose. Then anyone in the channel can tag @Claude, hand over a task in plain language, and carry on with their own work. It breaks the task into stages, works through them with the tools it has, and replies in a thread with what it produced. It launched in beta for Claude Enterprise and Team customers, and replaces the older "Claude in Slack" app.

The four things that make it a teammate, not a chatbot

What separates Claude Tag from a normal chatbot comes down to four behaviours.

What Claude Tag actually changes in day-to-day work

The easiest way to understand it is the kind of jobs people are handing it. A few that map onto almost any team:

• Summarise a long thread. Tag @Claude on a 60-message channel and ask for the decisions and open actions, with owners, instead of reading it all back.
• Chase a stalled task. Ask it to follow up on threads that have gone quiet, so things stop slipping through the cracks.
• Triage support tickets. Point it at the queue to draft replies grounded in your knowledge base and flag the ones that need a human.
• Pull a metric or a number. Ask it to chase down a figure from a connected tool and post it in the channel, rather than waiting on the one person who knows the dashboard.
• Draft from context. Because it already follows the channel, it can draft the update, the brief or the first version without a long set-up prompt.

Here is the same job, before and after.

This is agentic AI in action

Step back and Claude Tag is a clear, real-world example of the shift from prompting to working with agents, the move we cover in our guide to the agentic AI apprenticeship. For a couple of years, using AI at work mostly meant opening a separate chat window and copying answers back into your real tools. Claude Tag closes that gap: the AI is in the channel, it plans and executes multi-step work, and it only comes back to a human for the parts that need judgement.

That changes the question employers should be asking. A year ago it was "can our people write a decent prompt?" Now it is "can our people delegate to, supervise and govern an AI teammate?" Those are different, more valuable skills, and they will apply to whatever agent lands in your tools next, Microsoft is pushing Copilot agents and Google is doing the same with Gemini.

The bit that decides whether it works

This is the heart of it, and it is worth saying plainly without a sales pitch attached. Most organisations will get the licence. Far fewer will get the capability. That is where the real advantage lies.

Rolling out Claude Tag, or Copilot, or any agent platform, is the easy part. The value comes from people who can break work down for an agent, hand it over cleanly, and check what comes back, the same gap we wrote about in adoption is not ROI. Give a powerful agent to a team that has not built those habits and you do not get leverage, you get confident-looking output nobody has verified. Closing that gap is a training question, which is exactly what the AI & Automation Practitioner Level 4 is built to do.

Governance: what good looks like

To Anthropic's credit, Claude Tag ships with real controls. Administrators choose which tools and data Claude can touch in which channels, set token-spend limits for the organisation and individual channels, and can view a full log of what Claude did and who asked for it. Memories stay scoped, so a Claude set up for sales will not pass knowledge to one set up for engineering.

But controls only help if someone uses them well. Good governance for an AI teammate looks like this:

• Scope tightly. Give @Claude only the channels, tools and data a task actually needs, and keep separate scopes for separate functions.
• Decide what never goes in. Write down the data that must not be shared with an agent, and which channels it is allowed in.
• Keep a human in the loop. Someone signs off before anything Claude produces leaves the building or affects a customer.
• Cap spend and watch it. Use the org and channel token limits so cost cannot run away.
• Review the log. Check what the agent did and who requested it, the way you would audit any shared account.
• Start small. Prove it in a private channel on low-risk work, then widen access.

Deciding all of that is a leadership judgement, not an IT toggle, and it matters even more in regulated and public-sector settings. It is what AI for Leaders & Managers (Level 4) and the AI Adoption & Governance unit are built to cover.

How to try it

If you are a Claude Enterprise or Team customer, Claude Tag is in beta now, and Anthropic's setup is four admin steps:

That gets you running. Whether it pays off is the capability question above, and that is the part worth investing in.