Slack’s old Claude connector dies Aug 3, replaced by always-on Claude Tag

Anthropic is killing off its existing Claude in Slack app on August 3, and swapping in a new always-on agent called Claude Tag. The key change is that Tag is not just a “connect and ask” bot. It joins Slack organizational instances “as a team member,” stays present in approved channels, and learns from what it is given access to over time, rather than requiring every request to start from scratch.

Here is the practical consequence for decision-makers: if you run Slack at an enterprise or across teams, you are not merely updating a feature. You are deciding whether an AI agent becomes a persistent coworker inside channels that administrators choose, with permissioned access to the tools, data, and codebases those channels point to. In channels where Claude Tag exists, users can tag “@Claude” to delegate tasks to the bot, and the bot can then perform actions and build context based on what flows through that channel. Anthropic describes Claude Tag as “the beginning of an evolution of Claude Code,” aiming to make Claude more proactive and better suited for working with a full team.

This is not the first Claude integration into Slack, but the old approach is being superseded. There’s already an existing Slack connector app for Claude, and Anthropic is deprecating it in favor of Tag. According to the Claude in Slack help page, the existing connector will leave service on August 3. Anthropic also says Enterprise and Teams customers are getting access to Tag beginning today, while Slack environment administrators using the prior app have 30 days to opt in to the Tag migration.

So what does “always-on agentic” actually mean in the day-to-day? Anthropic says Tag is shared across a channel, which makes it behave more like a collaborative teammate than a series of one-off responses. When multiple people interact with the same channel’s Tag, they are effectively interacting with the same Claude instance in that space. On top of that, Tag “learns the longer it’s in a channel,” so users do not have to re-explain the basics every time they chat. If Tag is permitted, a Claude Tag in one channel can also learn from other channels and data sources, which pushes the integration from local context into a more connected knowledge picture.

Anthropic also positions Tag as proactive beyond direct mentions. It can “keep you updated about whatever it thinks you might need to know” across channels and tools it has access to, as long as ambient behavior is toggled on. It can also schedule tasks for itself to autonomously pursue assignments “over hours or days.” Anthropic says its employees found that feature useful, describing a shift from doing work themselves to delegating tasks to many Claudes in parallel.

There’s a second-order operational signal hidden in that description: Tag is being used for work that looks like continuous throughput. Anthropic says it’s using Tag to write code, stating that 65 percent of the product team’s code is apparently written by Tag. It also says Tag can chase down product metrics and data, work through support tickets, and help find the root cause of tricky bugs. Even if you ignore the exact internal metric, the list matters because it maps Tag onto the workflows enterprises already struggle to scale: debugging, ticket triage, metrics hunting, and the back-and-forth that happens when humans are the interface to every system.

Now layer in incentives and cost optics. Anthropic says it is offering “a hefty volume of credits” for Enterprise or Teams plans with more than 10 total seats, as long as those customers migrate before the credits expire on September 1. Support page details are specific: Enterprise customers adding Tag to Slack get $25,000 worth of credits, while Teams customers get $2,500. Those credits apply to Tag usage in Slack, but not to direct messages with Claude, which are billed to individual seats. Token use and metered costs have been a hot topic recently, especially since Anthropic moved away from bundled usage pricing to metered pricing that surprised customers with large bills. The company did not respond to questions for this story, and it is not clear how far the credits will stretch. For CFOs and board members, that uncertainty is the point: pilots and rollouts may look cheap until usage patterns get “always-on.”

Finally, there is the governance and risk side, which executives cannot outsource to IT. Using Tag means handing business data over to an AI model that lives on Anthropic’s servers. The source explicitly flags caveat emptor for channels where sensitive information may be discussed and ingested for contextual learning. At the same time, the article notes that Slack itself is already shunting potentially sensitive business data into the cloud, including through third parties like Salesforce. Translation: this is not an “AI or no AI” purity test. It is a “where does the data go, under what permissions, and what does the model do with it over time” question.

This migration sets a tone for the enterprise AI era: integrations are moving from chat overlays to persistent agents embedded in the collaboration layer. If you are responsible for platform risk, budgeting, or productivity tooling, Claude Tag is a live case study in how fast the agent paradigm is arriving at the systems where work actually happens. The August 3 deprecation date is the deadline, but the real stakes are earlier: permissions, channel scope, ambient behavior settings, and cost control will determine whether Tag becomes a strategic automation partner or a governance headache that shows up in the monthly numbers.