Claude Agents Stop Chatting and Start Owning Work
Claude agents in Slack are Anthropic’s new task-owning AI coworkers that live inside collaboration channels, accept delegated work via @Claude tags, use approved tools and data, and then post the finished output back into the thread, shifting AI from side-panel chat to embedded enterprise workflow automation that participates where teams already coordinate projects and tasks. This is not a minor interface tweak; it is a signal that agentic AI is moving into the same space where managers assign owners and track progress. The product, called Claude Tag, went live on 23 June in beta for Claude Enterprise and Team customers and runs on Claude Opus 4.8, replacing the older Claude in Slack app that will retire on 3 August. Anthropic is betting that the next phase of productivity is not asking AI for answers, but handing it responsibility.
Inside Claude Tag in Slack: Multiplayer Memory and Asynchronous Execution
Claude agents in Slack matter because they behave less like chatbots and more like shared project participants. Tag @Claude in a Slack channel and it takes on the job itself, working through the steps with the tools it can reach and posting the result back to the thread. One Claude serves everyone in the channel, so the context of what it is doing belongs to the team, not a single user, which closes the handoff gap when someone logs off or goes on leave. As the agent follows a channel, it builds memory of the team’s work, vocabulary, and decisions, and with administrator permission can pull context from other channels and connected data sources. Crucially, you can set Claude a task and return to your own priorities while it works, even scheduling work over hours or days before it reports completion, which shifts enterprise workflow automation from "tracking tasks" to "executing them".
Autonomous Agents Raise Governance Stakes: Limits, Tokens and Audit Trails
The autonomy that makes Claude agents useful also makes agentic AI governance non‑negotiable. Claude Tag can run in an ambient mode where it watches channels, surfaces information it judges relevant, and pursues threads or tasks that have gone quiet, a clear step up in initiative that Anthropic recommends teams keep disabled until they understand its behavior. Setting up Claude Tag is explicitly an administrator’s job: an admin pairs it with a workspace, connects the tools and data it can reach, sets spending limits and chooses which channels it runs in. Each Claude identity is scoped so departmental agents cannot bleed context into other areas, which is essential when legal, engineering and HR hold very different sensitivities. Admins can view, edit and delete what Claude remembers from a central audit console, and Slack conversations follow existing retention policies, with data removed from Anthropic’s systems within 30 days if the integration is disconnected. This is the minimum bar for compliance tracking when an AI can read and act in live collaboration channels.
From Slack to Microsoft Teams: Platform Politics and IT Control
Anthropic is reportedly preparing a Claude agent for Microsoft Teams, putting AI agents for Microsoft Teams directly inside the daily collaboration layer of Microsoft 365 even as Microsoft builds its own Teams‑based AI coworker and delegated Office tools. There is no confirmed launch date or announced Teams version, but the Slack beta already shows the model Microsoft tenants will expect: channel tagging, shared organizational identity, approved tools and administrator controls. In Slack, administrators can define channel access, tool access and token‑spend limits, with channel work billed to the organization while direct messages run on an individual Claude account. The same questions will be sharper in Teams, where collaboration channels can include customer data, source code, sales discussions and personnel information, so any AI agents Microsoft Teams supports must match the governance, billing and retention guarantees already offered through Claude in Microsoft Foundry and its Agent Service for multi‑step planning, tool use and task execution across enterprise systems.
Enterprise Readiness: Workflow Gains Demand Hard Governance Lines
Anthropic’s own adoption of Claude Tag is a telling signal: an internal version now writes 65% of the product team’s code, including much of what built Claude Tag itself. Platforms have spent this year moving from task tracking toward coordinated AI‑assisted work, and Claude’s channel agents arrive in the middle of a direct fight over who owns the Slack or Teams channel as the place for coordinated work. For IT and business leaders, the lesson is clear: workplace agent deployment cannot be an experimental side project. It demands upfront decisions on channel permissions, tool access, spending limits, and compliance tracking, because collaboration channels often contain customer records, source code and personnel discussions. Anthropic’s introductory credits of USD 25,000 (approx. RM115000) for eligible Enterprise organizations and USD 2,500 (approx. RM11500) for Team customers show how aggressively vendors want teams to test this model, but the continuous, asynchronous monitoring that powers productivity also changes token consumption and raises switching costs over time. Enterprises should adopt Claude agents deliberately, treating them as new coworkers whose contracts and boundaries must be clearly defined.
