On June 23, Anthropic shipped a coworker. Claude Tag puts a single Claude inside Slack as a shared teammate: anyone types @Claude in a channel and the whole team is working with the same instance, one that remembers the project, breaks a job into steps, runs them while you sleep, and in ambient mode speaks up before it is asked. Anthropic says this exact setup already writes 65 percent of the code on its own product team. The product is the best of its kind. The address is the story: Claude Tag runs inside Slack, and Slack belongs to Salesforce, which sells a competing coworker in the same channel.
Claude Tag works, and it is good. The harder question is whether an AI teammate can build a lasting advantage on a surface a rival owns. We scored ten credible alternatives across the things that actually decide enterprise adoption, and the answer is sharp: the strongest product on the board is not the one most likely to win.
What Anthropic actually shipped
Strip Claude Tag to what is new. It is one Claude per workspace, not one per person: a shared identity the whole company talks to, so a half-finished task can pass from you to a teammate without re-explaining it to a fresh chat. It remembers. The context it builds over weeks means it already knows the codebase conventions, the naming, the recent decisions, with no re-brief. It works asynchronously: hand it a task and it decomposes the job into stages, runs the tools it is allowed to run, and posts the result back to the thread hours later. And it is ambient: in that mode it watches the channels it can see, follows up on threads, and surfaces what it judges relevant before anyone asks.
Two guardrails make that safe enough to sell. Admins decide exactly which channels, tools, and data each Claude can reach. And memory is walled by scope, so the Claude in a legal channel does not seed what it learns into engineering. That scoping is what separates an ambient coworker from a leak.
It is in beta for Claude Enterprise and Team customers, with launch credits to trial it company-wide. It also retires something: the old Claude in Slack app shuts down on August 3, with a 30-day window to migrate. Anthropic is not bolting a feature onto the old assistant. It is swapping a chatbot for a worker and moving everyone across.
Claude in Slack (retires Aug 3)
The conversation
- Identity: one private assistant per person
- Memory: forgets between chats
- Work: answers only while you wait
- Initiative: none. It speaks when spoken to
Claude Tag
The channel
- Identity: one shared Claude per workspace
- Memory: persistent context of the team's work
- Work: runs tasks async, reports back hours later
- Initiative: ambient. It acts before it is asked
65%
Share of its own product team's code Anthropic says Claude now writes. The vendor is its own first heavy user.
Aug 3, 2026
The old Claude in Slack app retires. Admins get 30 days to migrate to Claude Tag.
$27.7B
What Salesforce paid for Slack in 2021. Claude Tag's channel is a competitor's asset.
The model is rentable. The memory is not.
The reason this matters past launch week is where the durable value sits. Frontier models are converging and renting cheap; a buyer can swap the brain behind an agent in an afternoon. What does not swap easily is the accumulated context: who owns which account, why a decision went the way it did, the shape of the codebase, the team's norms. Claude Tag is built to hoard exactly that. The longer it sits in your channels, the more it knows, and the more expensive it is to pull out.
That is a real strategy, and it is the right one. The system that holds the richest, permission-aware memory of how your business runs owns something no model release erases. Anthropic running it to 65 percent of its own code is the tell that it believes this: the company is its own first power user, and it is staking the product on memory, not on being a smarter chatbot than the last one.
The plan has one flaw, and it is structural. The memory accrues inside Slack, and Slack is not Anthropic's. Every byte of company context Claude Tag earns, it earns on a surface owned by Salesforce, governed by Salesforce's permissions, ranked by Salesforce's product decisions, and sitting one click from Salesforce's own coworker.
Anthropic is playing on the referee's field
Salesforce bought Slack for about $27.7 billion in 2021, and it has spent this year making Slack the place enterprise AI happens. Its EVP for Slack, Rob Seaman, put the pitch in one line: "Slack is the only layer in the AI stack where teams work together." Salesforce sells its own agents into that layer, Slackbot and the Agentforce coworker, and at Claude Tag's launch it framed the lineup as a menu: Slackbot, Agentforce, or Claude Tag, all in the same channel. The platform owner is also a competitor, and it sets the rules of the room.
That is leverage no better model can answer. Salesforce decides the permission model Claude Tag runs under. It decides how prominent a third-party coworker is next to its own. It decides the API terms, the rate limits, and the data a Slack app may see. None of that is hostile today; the two companies shipped the integration together and Salesforce handed out trial credits. But the option to tighten any of it sits with Salesforce, not Anthropic. A strategy whose one durable asset compounds on a competitor's surface has left its most important lever in a rival's hand.
Ten ways into the same channel
To see how crowded that room is, we scored ten credible alternatives and adjacent competitors on the seven things that decide whether an AI coworker gets adopted: Slack and Teams presence, memory and company context, model neutrality, MCP and tool access, enterprise controls, the ability to take actions, and overall fit as a Claude Tag-style teammate. The scores are directional, read off public product docs, funding and customer signals, and operator chatter. They are not a vendor benchmark. They are a map of who stands where.
| # | Product | Slack | Mem | Neutral | MCP | Ent | Act | Fit | Σ/35 |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Salesforce + Agentforce | 5 | 4 | 2 | 4 | 5 | 5 | 5 | 30 |
| 2 | Glean | 4 | 5 | 4 | 5 | 5 | 4 | 4 | 31 |
| 3 | Dust | 4 | 4 | 5 | 5 | 4 | 4 | 5 | 31 |
| 4 | Microsoft Copilot | 5 | 4 | 2 | 4 | 5 | 5 | 4 | 29 |
| 5 | ServiceNow + Moveworks | 4 | 4 | 2 | 3 | 5 | 5 | 3 | 26 |
| 6 | Atlassian Rovo | 4 | 4 | 2 | 3 | 4 | 4 | 3 | 24 |
| 7 | Viktor | 5 | 4 | 3 | 3 | 3 | 5 | 5 | 28 |
| 8 | UnifyApps | 4 | 5 | 5 | 5 | 4 | 5 | 4 | 32 |
| 9 | Relevance AI | 4 | 3 | 4 | 3 | 3 | 5 | 3 | 25 |
| 10 | Zapier Agents | 3 | 3 | 4 | 4 | 3 | 5 | 3 | 25 |
Slack: Slack and Teams depth · Mem: memory and company context · Neutral: model neutrality · MCP: tools and MCP access · Ent: enterprise controls · Act: ability to act · Fit: fit as a Claude Tag-style teammate · Σ/35: total. Rank is credibility as a direct alternative, not the raw total. The two highlighted rows are the tell.
Read the table for one thing and it gives up the whole argument. The highest raw score belongs to UnifyApps at 32 out of 35, a genuinely strong model-neutral platform, and UnifyApps ranks eighth as a Claude Tag threat. The most credible direct alternative is Salesforce's own Slack-native stack, which scores 30 and ranks first. Capability did not set the order. Distribution did. Whoever already owns the channel, and the permission graph inside it, starts ahead of whoever merely built the better agent.
The independents prove the squeeze. Glean has the strongest standalone company-context platform, a $7.2 billion valuation and about $300 million in annual recurring revenue, and it still has to win distribution one connector at a time against tools that ship inside the surface. Dust is the closest thing to Claude Tag's model-neutral twin: model-agnostic by design, a $40 million Series B led by Sequoia, agents already running in Slack and Teams across 3,000-plus organizations. ServiceNow paid $2.85 billion for Moveworks to own the employee service front door, the same job an ambient Slack coworker quietly attacks. Microsoft Copilot needs no Slack at all; it owns Teams and M365 and wins by distribution the way Salesforce does. Under every row the pattern repeats: the surface owners start with the room, and everyone else has to earn attention inside it.
Own the context, not the channel
If distribution decides the channel, the move is to stop fighting for the channel. The scan points at one position no incumbent will take: a portable company-memory layer that belongs to the customer, not to any platform and not to any model. Permission-aware, model-neutral, exportable, auditable, and able to act, working across Slack, Teams, email, docs, tickets, and code at once, so no single owner can revoke it or rank it down.
The prize is not a smarter agent in one app. It is the memory of how a company works, held somewhere no platform owner can switch off.
The shape of that layer is not a mystery. It is eight decisions, and the only hard rule is that none of them is allowed to depend on a single vendor.
This is the position Anthropic can half-take and half-cannot. Claude Tag is permission-aware and admin-scoped, which is the first column. It is bound to Claude, which forfeits the routing column. And it is bound to Slack, which forfeits the surfaces column. The first is a feature. The second two are the ceiling. Whoever ships the same memory without the lock-in sells to every company that does not want its institutional knowledge living inside one vendor's chat product.
Our Call
By June 2027, Claude Tag is no longer Slack-only. Anthropic ships it on at least one ambient surface its rivals do not own, its own desktop or web client, or email, rather than living only inside Slack and Teams.
The case: Claude Tag's one durable asset is the company memory it accrues, and right now it accrues on Slack, which Salesforce owns and where Salesforce sells a competing coworker. A teammate that can be ranked, repriced, or out-placed by the company that runs its only home has every reason to grow a second home it controls. Anthropic already owns the client, the Claude.ai web app and the desktop app, and it already lives in this product daily at 65 percent of its own code. Daily users notice platform risk first, and they are the ones who decide what gets built next.
What proves us wrong: if, by June 25, 2027, Claude Tag still ships only inside third-party chat platforms, Slack and at most Teams, with no Anthropic-owned ambient surface, the call is wrong. A Teams launch alone does not count. Teams is Microsoft's room, another rival's, and the point of the call is a surface Anthropic itself controls.
Settles: June 25, 2027.
Frequently asked questions
What is Claude Tag?
Claude Tag is Anthropic's AI teammate for Slack, launched June 23, 2026. You type @Claude in any channel and the whole team works with one shared Claude that remembers the team's context, breaks tasks into steps and runs them on its own, and in ambient mode follows threads and surfaces information without being asked. It is in beta for Claude Enterprise and Team plans and replaces the older Claude in Slack app, which retires August 3, 2026.
How is Claude Tag different from the old Claude in Slack app?
The old app gave each person a private assistant that answered when asked and forgot afterward. Claude Tag is one shared identity for a channel, with persistent memory of the team's work, asynchronous task execution, and an ambient mode that can act on its own. Admins scope which channels, tools, and data each Claude can touch, and memory stays walled per scope, so a legal channel's context does not bleed into engineering.
What are the best Claude Tag alternatives?
The closest direct alternatives live where Slack and Teams already are. Salesforce's own Slack-native agents, Slackbot and Agentforce, have the deepest distribution. Glean is the strongest independent company-context platform, at a $7.2 billion valuation and roughly $300 million in annual recurring revenue. Dust is the closest model-neutral teammate, model-agnostic with a $40 million Series B. Microsoft Copilot owns Teams and M365. ServiceNow plus Moveworks, a $2.85 billion deal, owns the employee service front door.
Does Anthropic own Slack?
No. Salesforce owns Slack, which it bought for about $27.7 billion in 2021. Claude Tag runs as a third-party app inside Slack, alongside Salesforce's own AI coworkers, under Salesforce's permission model. That platform dependency, not the model's quality, is the main strategic risk to Claude Tag.
References and research base
- Claude Tag launch (June 23, 2026): the product, the four properties, plan availability, and admin scoping, via TechCrunch, VentureBeat, and Fortune.
- The 65 percent internal-code figure, Anthropic's own claim about its product team's use of Claude, reported by The Decoder.
- The Salesforce and Slack relationship, the three-way choice of Slackbot, Agentforce, and Claude Tag, and the Rob Seaman quote, via Salesforce Ben.
- Glean's $7.2 billion valuation and roughly $300 million ARR: Glean and CNBC.
- Dust's $40 million Series B led by Sequoia, model-agnostic design, and 3,000-plus organizations: Sifted and Tech.eu.
- ServiceNow's $2.85 billion acquisition of Moveworks: ServiceNow.
- Salesforce's acquisition of Slack for about $27.7 billion, completed in 2021: widely documented; see Salesforce's own corporate record and contemporaneous reporting.
Source-quality note
The product facts (what Claude Tag does, the launch and retirement dates, plan availability) and the funding figures here are reported, drawn from the launch coverage and the companies' own announcements, dated June 2026 unless noted. The 65 percent figure is Anthropic's own claim about its internal use, reported by multiple outlets, and should be read as the vendor's number. The seven-dimension scorecard is our directional read of a fast-moving market, not a vendor-supplied benchmark; treat the ranks as argument, not measurement. The thesis, that distribution beats capability here and that Claude Tag's memory advantage compounds on a surface Salesforce controls, is this publication's analysis.