Claude Tag in Slack: when AI becomes a permanent colleague
Anthropic is launching Claude Tag, an AI integrated directly into Slack that progressively learns your company’s context, culture, and workflows. On the surface: it’s a productivity assistant. In reality: it’s a permanent sensor of your internal data.
Here’s what actually happens. Every Slack message containing @Claude Tag travels through Anthropic’s servers. The AI accumulates detailed knowledge of your projects, clients, internal tensions, business decisions. It’s technically groundbreaking—Claude truly understands your context, not just your isolated question.
But this “understanding” creates two concrete problems:
First, privacy. Even with contractual clauses, your sensitive data travels, gets stored, potentially indexed. For an SMB with trade secrets or client data, that’s real risk. Anthropic claims to respect confidentiality, but their business model relies on exploiting this data to improve Claude.
Second, dependency. The more Claude Tag knows your business, the more indispensable it becomes. You’re creating massive switching costs. It’s a classic tactic: low cost initially, then gradual lock-in.
The real story: Claude Tag isn’t cheap out of kindness. It’s an investment to become the AI layer of your business processes.
What this means for your business
For an SMB, the question isn’t “Is Claude Tag good?” but “Can we afford not to use it?”
If your competitors adopt it, you lose speed. If you adopt it, you accept technological dependency and implicitly accept that your data flows to Anthropic.
Our advice: test in limited scope (one channel only, no strategic conversations). Clearly document which sensitive data should never pass through Claude. Negotiate with Anthropic on data retention and usage—don’t sign standard terms. Finally, keep a backup plan: another AI assistant, or gradually shift to self-hosted models if dependency becomes too heavy.
In brief
Accessibility vs. saturation: the AI app market fragments
Building an AI app was hard two years ago. It’s now doable in 48 hours with no-code. Result: the market explodes in quantity but collapses in quality. For an SMB, the real challenge isn’t creating an AI app anymore—it’s making it different and profitable against a thousand identical competitors.
Verifying AI without spending 8 hours: the real SMB bottleneck
AI does quick work, but its hallucinations are expensive. SMBs are looking for efficient fact-checking methods without destroying ROI. Currently: no standard exists. Everyone invents their own process. It’s a massive hidden cost that nobody quantifies.
Fika Jobs: AI video agent recruiting (raised $4M)
Fika Jobs blends TikTok and LinkedIn with AI agents conducting video interviews. For a fast-growing SMB that can’t afford a senior HR director, it’s tempting. Risk: the AI screens out good candidates because they’re not “charismatic on camera.” Worth testing carefully.
MoEngage: individualized AI agents for each customer (marketing)
The approach: a dedicated AI agent per customer for micro-personalization. Sounds revolutionary but demands significant data and computing power. Only relevant for SMBs with mature customer bases and clean data. Otherwise, it’s just spam with machine learning on top.
Get The AI Brief in your inbox
3x per week, the essentials of AI decoded for business leaders.