Key takeaways: Claude Tag is a strong first-party Slack agent for Claude Team and Enterprise customers. It offers shared channel work, channel/workspace memory, routines, central spend limits, and an organization audit view. Bob's advantage is the complete operating model around each assignment: 3,200+ app connections, per-person access to personal systems, person/channel/company memory with provenance and correction, email/webhook/inbox triggers, human approval before sensitive writes, mid-run steering, finished files, and a cost receipt on every completed run. Two serious products, with Bob built for broader operational work.
What Claude Tag is
Claude Tag is Anthropic's own agent inside Slack. You write to @Claude in a channel, a thread, or a DM, and it works there. It has been in public beta since June 23, 2026, and Anthropic lists it as available on Claude Team and Enterprise plans.
Setup is one job for one person. An owner in the Claude organization provisions an agent identity and attaches credentials to a channel, a workspace, or the whole organization. Nobody else installs anything: Anthropic's own getting-started page says there is nothing for an end user to install or configure, and that if @Claude is in your channel you can use it. On Team and Enterprise plans, channel work carries no per-seat charge and draws instead on an organization usage balance an owner funds.
What it is built for shows in the shape of it. A thread is one shared session and anyone in the channel can steer it. It keeps memory at channel and workspace scope, and admins can review, edit, or delete that memory. It can follow up proactively and run standing work. Anthropic also documents organization and per-channel spend limits plus an Audit view covering one-time tasks, scheduled tasks, and network calls made through Agent Identity. If your company already runs on Claude and wants shared-channel work under one centrally governed identity, that is a strong fit.
What Bob is
Bob is not a better chat window. Bob is a hire.
The belief underneath the product is that what teams are short of is not answers. It is finished work. And software only finishes work if it can reach the systems the work actually lives in, act under the right person's access, and be held to account for what it did afterwards. Every line below follows from that.
So Bob is built like a colleague. You @mention or DM him in Slack and he works across your connected apps in one pass. He connects to 3,200+ apps, and personal ones connect per person, so a request runs under your access and not a shared login's. Work can reach him from a sentence-based schedule, a webhook, an inbound email address, or an inbox watch. He reads images, PDFs, and spreadsheets and returns finished decks and workbooks. He remembers company, channel, and person context, shows who taught him a fact, and corrects stale memory. A teammate can steer or interrupt a live run. Before a sensitive external write he stops and asks. Every completed run ends with a receipt: credits, duration, and tool-call count. He works in Slack today, with Microsoft Teams in early access.
Coverage, side by side
Each row is a question a buyer asks anyway. The Claude Tag column says what Anthropic's public documentation said on July 28, 2026, and where the docs do not provide a comparable number or behavior, this table says so rather than guessing.
| HeyBob | Claude Tag | |
|---|---|---|
| Where does it work? | Slack today, with a Microsoft Teams surface in early access. | Slack. Anthropic's product page lists Microsoft Teams as coming soon, with a waitlist and no announced date. |
| Who can buy it? | Any Slack workspace. $100 in credits to start, no card. | Claude Team and Enterprise plans. Anthropic states it is not available on Free, Pro, or Max, or for third-party deployments, and that it is in public beta. |
| How many apps can it reach? | 3,200+ apps in a browsable catalog, plus API, MCP, and browser paths for the rest. | Owners connect selected tools and repositories and can add HTTP APIs or MCP servers. Anthropic does not publish a directly comparable app-directory count. |
| Whose access does it use? | Personal apps default to the asking person's own connection, so your query runs against your row in the CRM and your inbox, under your role. | Shared. Anthropic's docs say personal connectors apply only in DMs, never in channels, and that a connection is shared by everyone the scope covers, so anyone in those channels can act with it. |
| What can start a job? | An @mention or DM, a recurring schedule, an incoming webhook, a dedicated inbound email address, or a watched inbox event. | An @mention or DM plus standing or proactive work. Anthropic's overview documents scheduled follow-ups and check-ins; it does not document a general inbound webhook or dedicated inbound email trigger. |
| What can you hand it? | Images, PDFs, and spreadsheets read in the thread, and finished files come back the same way, including workbooks and decks. | Images and screenshots, yes. Slack canvases, no. Pasted links only if an admin allowed that site for the channel. PDFs are not covered either way in the documentation reviewed. |
| What does it remember? | Person, channel, and workspace scopes. Memory shows who taught Bob a fact and when, can be listed, edited, and deleted in Slack, and replaces stale facts when corrected. | Channel and workspace memory. Anthropic says admins can review, edit, and delete it, and that memory respects channel boundaries. |
| What happens before it writes? | Sensitive external writes pause in the thread with the app and action named plus Approve, Always allow, and Deny. Standing permission is narrowly bound and revocable in Slack. | Owners govern credentials and repository access by organization, workspace, and private channel. Anthropic's current overview does not describe a comparable in-thread pre-write approval card, so we do not assume one. |
| What can you see afterwards? | Tool calls, encrypted arguments and results, approvals and denials under one run, plus a receipt showing credits, duration, and tool-call count. | An organization Audit view lists one-time and scheduled tasks and Agent Identity network calls. Posts, commits, and pull requests are also attributed in their destination tools and link back to Slack. |
| How do you pay? | Credits you choose, from $99/month. Larger jobs can show an estimate, completed runs show a receipt, the workspace has a hard cap, unused credits roll over, connection runs are refunded, and Bob-caused reruns are free. | Channel work draws on organization usage. Owners can set organization-wide and per-channel hard limits, receive threshold alerts, and review per-channel usage. Direct messages bill the person's Claude account. |
What Bob connects to
This is the question that decides whether any of the rest matters. An AI employee that cannot open the system the work lives in sends you back to copying and pasting, and the whole promise collapses. So the answer has to be boring and complete.
Bob reaches 3,200+ apps. Connecting one is a sentence in the channel and a sign-in: say "connect HubSpot", Bob posts a Connect button, you sign in once, and every run after that can use it. Gmail, Google Calendar, HubSpot, and Linear connect per person, so Bob answers out of your inbox and your pipeline rather than a shared service account's.
| What your team runs | Apps Bob works in (examples) |
|---|---|
| CRM and pipeline | HubSpot, Salesforce, Pipedrive |
| Inbox and calendar | Gmail, Google Calendar, Microsoft Outlook |
| Docs and spreadsheets | Google Sheets, Google Drive, Notion, Microsoft Excel |
| Tickets and projects | Jira, Linear, Asana, ClickUp, monday |
| Support queues | Zendesk, Intercom, Front |
| Money and billing | Stripe, QuickBooks, Xero |
| Storefront | Shopify, WooCommerce, BigCommerce |
| Code and incidents | GitHub, GitLab, Sentry |
| Meetings and signatures | Zoom, Google Meet, Docusign, Calendly |
The catalog is the shortcut, not the ceiling. If your tool is not in it, Bob can work through its API, through an MCP server, or through the same browser your team uses, and for a core internal system we will build the connection. The full directory is on the integrations page.
Pick Claude Tag if your organization is already on Claude Team or Enterprise, the work is team work in shared channels, and you want one identity your owner provisions and governs centrally rather than each person connecting their own accounts.
Pick HeyBob if you want 3,200+ connections, each person acting under their own access, work that can start from email, an inbox, or a webhook, person/channel/company memory with provenance, a human gate before sensitive writes, finished files, and a per-run record of what the task cost.
If you are here because Anthropic's original Claude in Slack app is being switched over, we wrote about what that means and what to do about it.