Summary of Key Points
This article focuses on Claude Tag, an AI tool launched by Anthropic that can join Slack channels as a “team member.” It represents the third phase in the evolution of AI products from “chat assistants” and “code assistants” to “AI coworkers.” Anthropic has already extensively utilized Tag within its operations (65% of the company’s code was generated by it), believing it has the potential to unlock a market worth trillions of dollars. However, the implementation of Tag faces challenges such as high costs and security concerns regarding access rights. In the future, it could transform the way companies work and the structure of the AI industry.
Detailed Analysis
1. Claude Tag: More Than Just a Tool – An “Always Available Colleague”
Claude Tag is essentially a digital employee that behaves just like a real person within the company’s Slack teams. It can view all conversations and documents in the team (depending on permissions), and you can assign tasks to it by @-ing it; it will also proactively look for things to do on its own.
- Differences from Previous AI Tools: ChatGPT is a “question-and-answer” machine, while Claude Code is a “single-person code assistant,” but Tag is a “multi-person collaboration partner.” It understands the context of the team (such as details of previous discussions) and can work together with everyone, even identifying issues on its own (e.g., monitoring abnormal data or reminding people about unfinished tasks).
- How to Use It: For example, if a bug arises online, it will initiate monitoring, locate the problem, write the fix code, and then seek approval from the responsible person. New employees can ask HR questions by @-ing Tag directly without having to contact real people. It can also help you filter through important information from dozens or even hundreds of Slack channels and generate summaries.
2. Why Is Tag So Useful? Three “Superpowers”
Anthropic claims that Tag is effective because it solves three key problems:
- Long-Term Autonomous Work: Ordinary AI models can only work for a few minutes at a time, but Tag can operate stably for several hours or even connect multiple tasks over months (e.g., reminding itself to check experimental data next week). This transformation is like moving from a “temporary worker” to a “full-time employee” capable of handling long-term projects.
- Memory and Experience Learning: Tag has three levels of memory: the conversations related to the current task, the long-term rules of the channels, and the company-wide knowledge base. It doesn’t just memorize things mechanically but can abstract and apply experience (e.g., remembering your design preferences to help you improve future solutions).
- High Emotional Intelligence: It knows when to offer assistance (e.g., when someone raises an issue that no one is addressing) and when to remain silent (it doesn’t interrupt unnecessarily). It can even use emojis, making communication more natural.
3. What Tasks Can Tag Assist With?
- Short-Term: Routine Tasks: It can handle basic tasks such as those performed by junior operators, administrators, or engineers—writing code (65% of Anthropic’s internal code was generated by it), handling customer feedback, onboarding new employees, and checking data.
- **Long-Term: Becoming the Company’s “Brain”:
- Shared Decision-Making: It can integrate decisions from all teams (product, engineering, marketing) and alert employees to potential blind spots (e.g., “Someone just changed this feature; you need to adjust your plan”).
- Digital Twin: After working with you for a while, it will start to mimic your way of thinking and suggest improvements or initiate discussions on your behalf.
- End-to-End Project Management: It can take full responsibility for managing channel retention rates, from analyzing data, identifying issues, modifying code, to monitoring results—all without the need for multiple people to collaborate.
4. Challenges to Implementation
Despite its usefulness, Tag has not yet been widely adopted due to two main barriers:
- High Costs: Small companies cannot afford it; a team of 20 people using Tag could cost tens of thousands of dollars per month. For example, analyzing an issue with the website might cost $30, and conducting a thorough research project could cost $100. This is because Tag needs to process large amounts of context during each use, which is resource-intensive, and it also requires integration with many tools (such as GitHub and Notion), further increasing costs.
- Security and Permissions: To be effective, Tag needs access to significant company data and system permissions, but current security measures are not 100% reliable. Even if it’s 99% secure, there is still a risk of leaking confidential information. Therefore, only small companies with simpler architectures and higher levels of transparency dare to use it, while larger companies are hesitant to grant such extensive access.
5. The Future of Tag: Will It Change the AI Industry?
The emergence of Tag could lead to several significant changes:
- Opening Up a Trillion-Dollar Market: AI will transition from being a tool to becoming an employee, covering a much larger segment of the workforce ( outnumbering programmers), potentially expanding the market from tens of billions to trillions of dollars.
- Anthropic’s Growth Potential: Tag could shift the way companies use AI from relying on “human-initiated calls” to “AI-proactive tasks,” potentially shifting budget allocations from tools to human resources, as using Tag becomes more cost-effective.
- Competitive Advantages for Model Companies: Tag will hold valuable information about company operations (such as ongoing tasks, permissions, and dependencies). Replacing it with another AI model would be like replacing a team of employees, resulting in high migration costs. This could give Anthropic a competitive advantage over open-source models.
- Control Over AI: Both overseas model companies (like Anthropic) and domestic giants (such as ByteDance and Tencent) are competing for control over AI coworkers. Overseas companies rely on model capabilities, while domestic companies focus on collaborative ecosystems (e.g., combining tools like Lark and DouBao).
Conclusion
Claude Tag marks a crucial step in the evolution of AI from a tool to a collaborator. Its effectiveness has been proven internally, but cost and security concerns are the main obstacles at present. If these issues are resolved, it could become as disruptive as Claude Code last year, fundamentally changing the way white-collar workers perform their jobs. For companies, this could mean increased efficiency or the potential replacement of human employees. For AI companies, it represents new growth opportunities or a competitive battleground. This development deserves continued attention.