Split Personality: Getting your head around Claude Tag

This week Anthropic launched Claude Tag - an improved agent that your team can chat with in Slack.

It was easy to scroll through the release and miss the implications of this change and what it means to your business - in this post I'll help you with that.

How did things used to work?

I'm speaking broadly about agents now, but generally, tools like Claude Code, or Cowork, or AI tools in your SaaS apps have been an extension of your own personality.

When you set them up you plugged them into your other tools using your user account and permissions.

When Claude Cowork accesses Google Drive or Hubspot it does so 'as you' and can only see and update the things that you could see or update if you went to that system yourself.Amazingly powerful, yet it is just an extension of you.

If you don't give the agent instructions, or permission to use the tools, nothing happens.

Claude Tag has its own identity

With Claude Tag, we move from Claude being an extension of your own persona, to it becoming independent in its own right.Claude Tag has its own persona, with its own tools

Your admin team (IT likely) will set Claude up with its own accounts in the source systems:

• Email/Docs: create its own user account - claude@yourcompanyname.com
• SaaS (eg CRM, ERP, Data)- create an account with claude@yourcompanyname.com or a API service account

When thinking about what systems to connect consider how you onboard human employees:

• Knowledge and docs - intranet, Google Drive, OneDrive, Notion
• Code - Github
• Data - Snowflake, Tableau, BigQuery
• Go-to-market - Hubspot, Salesforce, Gong, DocuSign
• Calls - Granola
• Comms - Email, Calendar

You want dedicated accounts that are not linked to a human user or any other API key used for other purposes.

By giving Claude these new user accounts you can be very specific about what permissions to give it - read only? read/write? which folders/files should Claude have access to?

Having created all of these Claude Tag allows you to wrap these into an "Access Bundle" that you can give to your Claude agent.

How will your employees work with Claude Tag?

Until now interactions with Claude (whether in Claude.ai, Claude Code, or via Slack) were single player.

I have a question or a task, I ask Claude, Claude helps me, I can then choose what to do with that response.

Even when chatting with Claude in Slack, it was still 'my Claude' with 'my persona'.

Now your employees can tag Claude in any public channel, and Claude's new 'persona' with its own access to systems and tools can get to work.

Consider it the same as pinging a human colleague and asking them to help you with something - except this time Claude now has its own identity, systems, skills and memory to get on and do the work, keeping the channel informed with its progress.

What this means for your organisational planning

Now that you separate Claude's identity from that of your human employees you can start to thing of it as additional to your human team, not complementary.

If you had 100 people, you now have 101.

Except that actually you now have 200, or 2000... because Claude can do multiple things, for multiple people at the same time.

You will want someone to become Claude's manager, Claude's coach, Claude's trainer - feeding it knowledge, skills, operating procedures, best practice - ensuring that the user accounts you provided (email, docs, SaaS) compound the knowledge and experience Claude has.

Each Slack channel that has Claude in it can be personalised with its own memory, so even though you've just got your head around Claude having its own personality, you should start to consider you have many Claudes who are each individual. 🤯

Today the Claude app cannot be renamed, or given a new avatar, but I would expect that to come soon - here at Kowalah we run multiple agents built on Anthropic Managed Agents:

• Brian - our AE
• Krista - our Demand Gen lead
• Karolina - our project manager

They each have their own personality, skills, tool permissions and memory. Feedback from our human team is they love being able to speak to Brian about open deals, Krista about upcoming events, or Karolina to dig into a current project.

Each agent has its own instructions, set of tools and permissions which makes this work so well.

Claude Tag has most of this available at launch so I would expect multi-Claude personas as a natural next step.

Beware of the incoming Anthropic bill...

One final gotcha as you roll this out to your team. This is going to become highly valuable, highly addictive, and highly adopted by your team. All usage of Claude Tag goes by the API at API rates, not using individual subscriptions. If you're not careful in terms of:

• how you set up Claude Tag
• how you set spend limits
• the skills that you give to Claude to show how to use the right tools
• how your team sets up routines that run on a schedule

there's a very real risk of getting a bill that you were not expecting.Claude Tag includes settings to provide these restrictions, but beware giving people access to something so valuable and useful that you immediately have to restrict or remove Change enablement and a governed roll-out is the way.

For more information on how to set up Claude Tag as an administrator click here.

Or speak to anyone at Kowalah, and we're happy to help you through this