A friendly course on Claude Tag: the way to bring Claude into your team's Slack chats. No tech words needed. Every idea comes with a why.
Picture a playground. Your Slack is the playground. Your team is the kids. Claude Tag is a new friend who joins the game.
Type @Claude in a Slack message. Claude reads the thread and the channel around it, so you don't have to explain everything again.
A whole team can talk to Claude in one thread. Anyone can add a hint or fix direction.
Claude can work on longer jobs and on a schedule, like "every Friday at 4", even when you're away.
Tell Claude a rule once and it keeps it in that channel's "notes".
Claude is a fast, creative helper. It is not a judge of what is true. Like any AI, it can say wrong things in a very sure voice. That is called a hallucination. You bring the ideas and the checking. Claude brings the speed. You will practice this in Level 5.
Good to know: Claude Tag launched in beta in June 2026 for Team and Enterprise plans. Features can change, so check your admin's settings.
Where you tag Claude changes what it can see and who can see the answer. Tap a field.
Giving Claude a job works like a game of tag. Four moves. Tap each one.
If Claude starts running the wrong way, just reply in the same thread. Claude reads new replies while it works. No new @Claude needed.
Pick your team. Each idea has something to type, the reason it helps, and a creative twist to make it your own. Tap a prompt to copy it.
Claude does the work faster. You decide what is worth doing. Creativity is your part: the odd question, the new angle, the "what if". Tap a spark card to flip it.
Pick a team and a pain. We'll mix a prompt starter for you.
AI models sometimes make things up. They can invent a number, a quote, a link, or a rule, and sound totally sure. The fix is simple: people check. Claude drafts. You verify.
Whenever Claude hands you work, you tag yourself back in. Five checks, from quick to careful:
Claude wrote a summary. One line has nothing to back it up. Tap the line you would check first.
Try: "Now go through your answer. For each claim, find the exact quote in the source. If you can't, say you can't and remove the claim." This makes a weak claim visible. It still does not replace your own check. Anthropic's own guidance on reducing hallucinations suggests citations and comparing repeated answers.
When you ask for work, link the raw data, ticket, or contract. Then you and Claude are both looking at the same facts. That is why the four moves in Level 2 ask for a "verification method" in the brief.
Six quick questions. Each right answer earns a star.
Tick these off for your own workspace. Your ticks are saved in this browser.
Play the real game. You tag Claude, Claude runs, and when it needs you it tags you back.