AI Agents
Claude Tag vs Slackbot vs an Agentforce agent: which assistant belongs in Slack
A Slack workspace at a Salesforce customer can now hold three AI members that answer questions and take actions: Anthropic's Claude Tag, Slack's own Slackbot, and an Agentforce employee agent you built. They run on different plans, hold data in different places and act under different permissions. This is the comparison, and the one question that decides which you turn on for which job.
A support engineer types a question into a channel and three things could answer it. Slackbot, which is a member of the channel. Claude, if someone tagged @Claude. Or the IT agent the Salesforce team built in Agentforce and connected to Slack last quarter.
All three are real products at a Salesforce customer in September 2026, and the confusion between them is not the user’s fault. Each is sold as an AI teammate in Slack.
They differ in who runs them, what they know, what they can touch and who pays. Get the split right and each does a job the others cannot. Get it wrong and you have three assistants answering the same question three ways.
What each one is
Claude Tag is Anthropic’s product. Launched on June 23, it replaces the old Claude in Slack app with an always-on teammate you tag with @Claude. Each channel gets one shared Claude with a persistent memory and an organisational identity, so a colleague can pick up a thread where another left it.
It follows the channel, learns the work, and in ambient mode it flags items and follows up on things people forgot. Admins define which tools, information and channels it may reach, and identities are scoped per channel. Anthropic gives the example that a legal channel’s Claude cannot share memories with the engineering channel.
Claude Tag is in beta for Claude Enterprise and Claude Team customers.
Slackbot is Slack’s own assistant, and since the Claudeforce partnership it runs on Claude by default. It joins channels as a member on Business+ and Enterprise+ plans, remembers user preferences, runs reusable AI Skills, connects to outside tools through an MCP client, and can build live Salesforce panels inside Slack.
I covered the full 2026 feature set separately. In one line, Slackbot knows the channel and, increasingly, the Salesforce data behind it.
An Agentforce employee agent is one you built. It has topics and actions defined in Agentforce Builder, it acts on your org’s records through Flows, Apex and prompt templates, it runs under an agent user with permissions you assigned, and it is metered in Flex Credits.
Salesforce’s named agent Paige, for IT and HR requests, is the packaged version. I wrote up how these agents behave in Slack and what a named agent changes.
The comparison
| Claude Tag | Slackbot | Agentforce employee agent | |
|---|---|---|---|
| Who runs it | Anthropic | Slack | You, on Salesforce |
| Plan or licence | Claude Enterprise or Team, in beta | Slack Business+ or Enterprise+ | Agentforce, metered in Flex Credits |
| What it knows | The channels it may read, with persistent memory per identity | The channels it is in, user preferences, Salesforce data through Surfaces | Your org’s records, knowledge and Data 360, through grounding |
| What it can do | Tasks through connected tools, asynchronously over hours or days | AI Skills, MCP tool calls, build Surfaces | Named actions: create the ticket, reset the access, update the record |
| Whose permissions | Admin-defined tool and channel access per Claude identity | Slack permissions plus whatever the MCP connections allow | The agent user’s Salesforce permissions, per action |
| Where the data goes | Anthropic, with zero data retention promised and Enterprise Frontier Safeguards from the fall | Slack, with Claude as the model underneath | Salesforce, inside the trust boundary, with session traces in Data 360 |
| Audit trail | Slack history plus Anthropic-side controls | Slack history | Session tracing, Trust Layer audit, Salesforce logs |
Read the table by column and the shapes are different. Claude Tag is a generalist that learns a team. Slackbot is a workspace-wide assistant with a growing set of tools. An Agentforce agent is a specialist with a defined job and a permission set to match.
The question that decides it
Ask what the assistant is allowed to change, and in which system.
If the answer is nothing, and the job is thinking work, drafting, research and following up on a channel’s threads, Claude Tag is built for that. Its memory and its asynchronous mode are the features the other two do not have in the same form. It belongs to teams already on a Claude Enterprise or Team plan, which is the gating factor.
If the answer is a little, across many tools, for everyone, Slackbot is the default. It comes with the Slack plan most companies already pay for, it reaches Salesforce data through Surfaces, and its MCP client can be pointed at other systems. The governance cost is that the reach is defined by whichever connections an admin enables, so the inventory matters.
If the answer is a specific record in Salesforce, with a rule about who may change it, build the Agentforce agent. An IT request that resets a permission set, an HR request that opens a case, a sales question that updates an opportunity: these want a named action, a permission check per action and a trace of what was done. That is what the Agentforce agent gives you and the other two do not, because the other two act as a user with tools rather than as an agent with a permission set.
Where they overlap, and what to do about it
The overlap is answering questions from Salesforce data. Slackbot can surface a pipeline panel. Claude Tag, if given the Salesforce in Claude connection, can answer from CRM records. An Agentforce agent can be asked the same thing. Three answers from three assistants with three permission models is the failure mode.
The fix is a rule, not a product. Decide that Salesforce writes go through the Agentforce agent, that Salesforce reads inside Slack go through Surfaces, and that Claude Tag is for the channel’s own work. Then wire the Slack workflows so a Slackbot answer that needs an action hands to the agent rather than trying to act itself. That mirrors what Salesforce did with Coworker: the front door answers, the agent acts.
Skills are the other overlap. Slackbot has AI Skills, Coworker has AI Skills, and Anthropic’s Agent Skills format underlies the Claude side. They are three different things with one name. The governance question is the same for all three: who may create a skill that other people run, and who reviews it.
What I would turn on this quarter
For a company on Business+ or Enterprise+ with Salesforce, in this order.
Slackbot first, because it is already in the plan and the channel-membership decisions are the ones to make early. Then one Agentforce employee agent for the highest-volume internal request that ends in a record change, with Paige as the starting point if IT or HR is the queue. Claude Tag last, for the teams that already have Claude Enterprise seats and do knowledge work in channels, with its channel scope set narrowly.
Coding work is a fourth case with its own product. If engineers want agents in channels, that is Slack Code, and the same scoping questions apply with repositories instead of records.
Understanding the basics
What is the difference between Claude Tag and Slackbot?
Claude Tag is Anthropic’s AI teammate for Slack, tagged with @Claude, with a persistent memory per channel identity, available in beta on Claude Enterprise and Team plans. Slackbot is Slack’s own assistant, included on Business+ and Enterprise+ plans, running on Claude as its default model, with AI Skills, an MCP client and Salesforce Surfaces. One is a licensed Anthropic product you add; the other is part of Slack.
Can Claude Tag access Salesforce data?
Claude Tag reaches whatever tools and channels an admin grants it. Salesforce data reaches Claude through the Salesforce in Claude plugin under the Claudeforce partnership, which runs through the user’s Salesforce permissions. Whether a given Claude Tag identity has that connection is an admin decision.
Should I use Slackbot or an Agentforce agent for IT requests?
An Agentforce employee agent, or Salesforce’s packaged Paige agent, when the request ends in a change to a Salesforce record such as a case or a permission set. The agent runs named actions under a permission set and leaves a session trace. Slackbot is better for questions and for surfacing data, with a handoff to the agent when an action is needed.
Do these assistants cost extra?
Claude Tag requires a Claude Enterprise or Team plan. Slackbot AI is included on Business+ and Enterprise+ Slack plans. An Agentforce agent consumes Flex Credits per action under your Salesforce agreement.
One rule per assistant
Three AI members in one workspace is manageable if each has one sentence that says what it is for. Claude Tag does the channel’s thinking. Slackbot answers and surfaces. The Agentforce agent changes records. Write those three sentences down, set the permissions to match, and the question of which one answers stops being a coin toss.
If you want the three assistants in your Slack workspace given one permission model and one rule each before more of them arrive, talk to us, or check where your org stands with the Agentforce readiness assessment.
