Best practices for using @Claude

Claude now works alongside your team, under its own account, in the places you already work together. Tag @Claude with a task and it takes it on. The work happens in the thread where everyone can see it, and Claude keeps going after you've closed your laptop, coming back to you when it's done or when a decision needs you.

This guide covers how to set up your channels, write a task that goes well, and move from giving @Claude tasks to handing it work that runs on its own.

What makes this Claude different

You may have used Claude in a chat box, or have automations running in your workspace that each handle a specific job. This Claude is different in a few ways:

• The work keeps running after you leave: tag @Claude with a task and close your laptop. The work keeps going, and tasks you've put on a schedule run without anyone there. When you want to check in, reply in the thread from whatever device you're on. A teammate can do the same while you're out.

• It's collaborative from the start: unlike working in a private chat box, @Claude is in your workspace, ready to take the next step on tasks you're already discussing. Claude is in the thread with the whole team from the start, so anyone can redirect or add to the work. Give it the problems you're already working through together. Everyone can see the progress, and jump in to give direction.

• It works where your context already is: Claude reads the channel's history to understand your work, so most of the briefing you'd normally give to Claude is already done. When a specific doc, thread, or past decision matters, name it and Claude reads it from your workspace. In a channel that's devoted to one kind of work, Claude starts becoming a specialist.

• It learns the channel and gets better at the job: corrections and preferences you give to Claude stick. When a result comes back, tell it what you liked or what you'd change, and it saves that for the channel going forward. The Claude in your channel becomes more tuned to your preferences and gets better at your type of work over time.

Tips while using @Claude in your workspace

1. Learn what Claude has access to

In a channel, Claude works with the connections your admin set up (your docs, your code, your data, or whatever has been connected) plus the channel's own history. Anyone in the channel can continue a task someone else started, and the tools and context are shared.

In your DM, it's just you and Claude. Your personal connections like your calendar and email work there, and only there. In channels, Claude uses its own admin-managed accounts, so it's not possible for anything from your personal tools to end up in a channel by accident.

Memory follows the same boundaries. What Claude learns in a public channel is available across your workspace, but what it learns in a private channel or a DM stays in that space.

Connections to tools help expand what Claude can do. A couple of connections and Claude is answering questions; connect most of what your team uses and Claude is completing the work. If you're not sure what it can use where you are, ask it: "@Claude what can you access from this channel?" If something you need isn't connected, ask your admin to add it.

2. Work with Claude in public

Do team-relevant work in public channels rather than DMs, when you can. Claude works from the channel's history, so months of decisions, corrections, and examples are available to Claude before you've given it the first task. Working in public also helps beyond that channel: your teammates can find and learn what you did, and so can Claude when it's working elsewhere in the workspace.

A channel about one kind of work gives Claude a clear focus. The history and corrections in that channel are what Claude learns from, so it gets good at exactly that channel's work. You can stop pasting a certain doc in for context or starting with a long briefing.

3. Teach Claude when to respond

By default, Claude only responds when someone tags it. It doesn't act on its own unless you've told it to. Initially, this is helpful place to start: you directly tag Claude into the conversation with a task, and Claude works on it in the thread. Overtime, you can begin to give Claude more range over when and how it responds.

Set Claude's level of proactivity: you can change how active Claude is in a given channel by telling it in plain words. Direct Claude to "reply only when tagged" or "answer every question here." Busy feedback channels benefit from the latter—errors get corrected immediately, and no questions go unanswered. A planning channel might want Claude to be quiet except for a scheduled update. If proactive replies aren't available in your channel, your admin can turn them on.

Schedule tasks: when you find yourself tagging Claude for the same thing weekly, tell it to run that on a schedule instead. For example: "Post a digest of decisions and open items every Monday at 9am." Claude handles it automatically each week without anyone tagging it.

Have it follow up on its own: Tell Claude what to watch out for or monitor. Claude can check back on its own to bring updates when they matter. For example, Claude can watch for new information that might affect a decision or alert you about a forgotten action item.

Working with @Claude on a task

Step 1: Pick the task

Set yourself up for success by thinking through what you want, what Claude can access, and how you'll know if it worked.

• Shift from step-by-step thinking to outcome thinking. Picture delegating to someone who understands your workspace completely. What would you ask them to pull together or figure out?
• Consider the inputs your task needs and make sure Claude can reach them. If you're unsure what's available, ask Claude what it can access in this channel.
• Think about what success looks like. How will you be able to tell if Claude got it right? Think about ways you can ask Claude to verify the work, like linking sources or explain its reasoning.

For examples, see use cases of what @Claude can do.

Step 2: Write the request

Brief Claude the way you'd brief a capable new teammate. State your goal clearly and explain why it matters. Guide Claude to the right place, by mentioning the threads, docs, or tools you want it to focus on. Ask Claude to link its sources. Leave the steps to Claude. It can read the channel, search the workspace, and use whatever tools are connected.

Step 3: Let Claude work

Claude reacts to your message, posts a short note that it's started, and keeps a checklist updated in the thread. It keeps working without you, and if you're working in a channel, anyone in it can jump in to add context or adjust Claude's plan.

Editing a message you already sent has no effect; send a new reply instead. Each top-level message starts a new task for Claude. Keep separate tasks in different threads, but remember that Claude's channel memory applies everywhere.

Step 4: Review the result

Read Claude's work before you use it, in proportion to what's at stake: skim a summary, read carefully anything going to a customer or changing a system. When something's wrong, say what specifically and ask for the revision in the thread. For anything analytical, ask Claude to take a second pass looking for errors in its first answer. If the same miss keeps happening, open Claude's session to see the steps it took, then tell Claude what to do differently. Claude remembers the correction for the channel, so the next person doesn't hit it.

Giving @Claude more responsibility over time

To get started, tag @Claude in a channel where real work happens and ask it something a teammate would normally answer. If you want a low-stakes place to try things first, make a channel just for that (naming it “your name” and “Claude” works). And if you're not sure what to ask, ask Claude what it can help with where you are.

Once Claude is reliably doing a few things, the next step is widening its responsibility, in addition to one-off tasks. What that looks like in practice:

• Give Claude something to own: instead of running a fixed task, tell Claude to manage an ongoing responsibility. For instance, tell Claude, "You're responsible for keeping this channel's open questions answered. Check daily, answer what you can, tag whoever's right for the rest." Claude decides what needs doing each day instead of running a fixed script.
• Let Claude improve itself: at the end of the day, ask it to look back over the channel's threads and write down what it would do differently. Claude finds real improvements when it has time set aside to look; it just doesn't do that on its own unless you ask. After a miss, do the same thing and have Claude save what it learns.
• Ask Claude what's been missed: once in a while, ask Claude what's been raised repeatedly across the channels it's in that nobody's answered, or what looks stalled. Because it can read across its channels, it can see the gaps no single person is looking at.

What makes this possible

A few things run underneath all of the above, and they're worth knowing because they're what makes it safe to hand Claude real work.

• Claude has its own identity: in channels it works under its own account, not yours. Every action it takes is recorded under that account, so there's always an answer to what Claude did and who asked for it.

• Its access is set per channel: your admin decides what Claude can reach in each channel, and being added to a channel is what gives it reach there. That's why the same Claude can open pull requests in an engineering channel and only read the wiki in another. Security and data handling covers how its work is kept separate from everything else.

• Memory belongs to the channel: what Claude remembers is kept with the channel and workspace, not with any one person. Anyone on the team can ask Claude what it remembers there, add to it, or tell it to forget something that's out of date.

If you already use Claude somewhere else

Coming from claude.ai: You can stop collecting all the context yourself. You don't need to paste in the background or re-explain the project each time. Name the thread, the doc, or the channel, and let Claude read it there. And where an answer in chat might hand the work back to you, with @Claude, it can take the next step itself, like opening the draft PR, filing the ticket, or tagging in a teammate, and report back in the thread.

Coming from the earlier Claude app for Slack: @Claude works under its own account with the access your admin sets per channel, remembers corrections for the channel, and takes on multi-step tasks rather than single replies. The FAQ covers the differences, and the Help Center covers what happens to the earlier app.

Learn more

• How @Claude works: memory, identity, and what Claude can access in each place.
• Get started and Good habits: the basics, and how to write tasks that are easy to check.
• @Claude use cases: tasks teams are running today, with example prompts.
• Admin setup guide: connecting tools, scoping access per channel, spend controls.
• What is Claude Tag in the Help Center: plan availability, billing, and who can set it up. For other account questions, see the Help Center home.