The “company brain” just became a product. The real question is who owns it.

Claude Tag puts Claude inside your workspace conversations: you tag it and it works using the company’s tools and the shared context, keeping the channel’s and workspace’s memory instead of throwing it away at the end of a task[1]. Early coverage sums it up like this: it’s learning how your company works, one message at a time[2].

There’s also an ownership choice already stated: work in channels is attributed to the organization’s identity, work in a direct message runs on your personal account, and only an Owner can configure access and channels[1].

For many cases that’s perfectly fine. But it’s exactly the point where it’s worth stopping to think.

Memory alone isn’t the new part: models already remember something between sessions. The new part is that a whole team is now tagging the same Claude. And the moment five people work on the same assistant, shared operating state stops being a luxury and becomes a requirement. Without common context, the same Claude answers every colleague differently, because each one feeds it disconnected pieces of the truth.

One person can keep the plan in their head. A team can’t. You need a place where “how we work” is written down, shared, and re-read by everyone, people and AI.

An assistant becomes useful to a team only when it reads the same context the team reads. Otherwise it isn’t a teammate: it’s five different assistants with the same name.

We call “company brain” the set of things a team needs to know in order to work: decisions made, project status, open questions, rules, next steps. As long as it lives in people’s heads and in chats, it doesn’t scale. When you put it inside an AI, it scales. But at that point the question is no longer technical, it’s about ownership: is that brain yours, or is it rented?

Two axes change everything. The first: where the data lives, inside a vendor’s infrastructure or in a place you control. The second: how it’s governed, automatically and opaquely or explicitly and auditably. The combination of the two tells you who really owns your company brain.

It’s not a holy war: they’re two approaches with different trade-offs. Automatic vendor memory is convenient and zero-setup, but it’s opaque (you can’t see what it learned), it lives where the vendor decides, and if you switch tools you leave it behind. Context on files you own takes a bit of discipline, but it’s readable, versionable, auditable, and it moves with you.

For anyone in a regulated industry, handling sensitive data, or who knows tools will change, the difference isn’t a detail: it’s the difference between an opaque memory and governed work. It’s the same reasoning as our guide on AI governance for SMEs: you can’t govern what you can’t see.

We’re not theorizing this now: we’ve been building it for months, and it’s public. cowork-os is the open-source kit in which we package our method for people who work with the business: the folder structure, the instructions, the automations, and the work-state update rule, so the company’s context sits in one place, on your computer, readable by you and by Claude[3]. code-os is the same idea for people who write code: operating context and best practices on files, drop-in to any codebase[4]. Two faces of the same method: explicit operating context, on files that stay yours.

It’s the difference we care about. The AI others leave you in slides, we put where you actually work, on files you own. Claude Tag makes this method more relevant, not less: the more AI enters teams, the more it matters to have context that’s yours.

If you’re a small team, you already work entirely in Slack, and you want speed with no fuss, the automatic memory of a tool like Claude Tag can be more than enough. There’s nothing to configure and it works. Saying so is part of our method: if the vendor’s AI is enough for you, we won’t sell you complexity.

Owned, governed context matters when at least one of these is true: you operate in a regulated industry or with sensitive data, you need to know and prove what the AI learned, you work with multiple tools and don’t want to be locked into one, or you want the know-how to stay in the company even when people change. There, an opaque memory inside a vendor isn’t enough anymore.

You don’t need to switch tools to start. You need to know where you stand. Answer these five questions about your current AI assistant.

1. [1]Anthropic, “What is Claude Tag”, Claude Help Center. support.claude.com
2. [2]TechCrunch, “Anthropic’s Claude Tag is learning your company, one Slack message at a time”, June 23, 2026. techcrunch.com
3. [3]cowork-os, open-source repository. github.com
4. [4]code-os, open-source repository. github.com