Poly vs Claude Tag for a team sharing one Claude

Claude Tag is Anthropic's shared Claude for Slack: one Claude per channel that anyone in the channel can steer by replying in its thread, sold on Claude Team and Enterprise plans and set up by an organization owner. Poly (usepoly.co) is a browser room where a small team steers one AI, Claude or ChatGPT, with an approval before it acts, a receipt on every turn and a live preview of the app, and it needs no admin, no Slack and no plan beyond a free account. They share the idea that the agent should belong to the group. They differ on where it lives, who can set it up, and what the group can see and decide.

What Claude Tag does

In Claude Tag, "within a given Slack channel, there's one Claude that interacts with everyone," and "once a session is active in a thread, it belongs to everyone there." Anyone in the channel can redirect a running session by replying. Each thread gets an ephemeral sandbox running the same engine as Claude Code on the web, and code work usually lands as a draft pull request opened by the Claude GitHub App. Access is per channel through admin-provisioned service accounts; usage is billed from an organization balance an owner funds, with a spend limit. It is in public beta as of September 2026, Slack only, with Microsoft Teams on a waitlist. Anthropic says 65% of its own product team's code is created by its internal version.

Where Claude Tag is stronger

• It is where your company already talks. If the team lives in Slack on a Claude Team or Enterprise plan, there is nothing new to open.
• Anthropic runs it. First-party support, the Claude Code engine, and admin governance that an IT department recognizes.
• Code lands as pull requests for review in GitHub, which suits an engineering team with an existing review process.

Where Poly is stronger

• No admin, no plan, no Slack. Two people open a room in a browser and start. Every account begins with free credits and no card.
• A say before it acts. In Poly's approval modes the AI's proposed change is a plain-language card on every member's screen before it happens. In Tag, steering is a reply in the thread and progress is a checklist the session edits.
• You see the app, not just the thread. Members watch the running app as a live preview beside the conversation and can publish a snapshot of it.
• Receipts. Every turn names who ran it and what it cost. Tag draws from one org balance.
• Two labs. Members pick Claude or ChatGPT per turn in the same room. Tag is Claude only.
• Built for people who do not code. Tag's output is a pull request; Poly's is a running app with an undo button.
• A channel the AI cannot read. Poly's group chat and polls are hidden from the agent by design. In Tag, the thread is the session.

Which should you pick?

• Your company is on Claude Team or Enterprise, lives in Slack, and reviews code in GitHub: Claude Tag.
• A small team, a cofounder pair, or anyone without an org admin: Poly.
• You want to see and approve the app as it is built, or use ChatGPT models too: Poly.

If your problem is two people wanting to work inside one Claude Code session, note that this is still an open, unanswered request on Claude Code's own issue tracker (#60082); Tag answers it for Slack-based orgs, Poly for everyone else. The Claude Code collaboration guide covers every option.