Claude Tag：你在 Slack 中的主动队友

Claude Tag: Your Proactive Teammate in Slack

拉米斯·穆克塔 Lamis Mukta · AI Native Dev · 2026-07-07 · 约 58 分钟 · 原视频 ↗

本期速览 · Overview

Anthropic 的 Lhamu Dolma 介绍 Claude Tag，一个在 Slack 中工作的主动 AI 队友，具有持久性、记忆能力，并能主动发起对话和长时间完成任务。

Anthropic's Lhamu Dolma introduces Claude Tag, a proactive AI teammate that works in Slack, with persistence, memory, and the ability to initiate conversations and complete tasks over long periods.

要点 · TL;DR

Claude Tag 是一个在 Slack 中具有跨频道持久记忆的主动式队友。
Claude Tag is a proactive Slack teammate with persistent memory across channels.
智能体自主性每 4 个月翻倍，支持更长时间运行的任务。
Agent autonomy doubles every 4 months, enabling longer-running tasks.
为未来模型能力构建，而非当前限制。
Build for future model capabilities, not current limitations.

核心观点 · Key points

Claude Tag 是在 Slack 中具有持久性和跨频道记忆的主动式队友。
Claude Tag is a proactive teammate in Slack with persistence and memory across channels.
智能体自主性大约每 4 个月翻倍，使得更长时间运行的任务成为可能。
Agent autonomy doubles roughly every 4 months, enabling longer-running tasks.
为模型未来的能力而构建，而非其当前水平。
Build for where models will be in the future, not where they are today.
定义成功标准是关键；智能体可以通过测试和评估自我验证。
Defining success criteria is key; agents can self-verify with tests and evals.
Dreaming 通过审查智能体记录和记忆存储实现持续学习。
Dreaming enables continual learning by reviewing agent transcripts and memory stores.

反共识 · Contrarian takes

Slack 正成为智能体式编码的主要界面，而不仅仅是 IDE。
Slack is becoming a primary surface for agentic coding, not just the IDE.
随着模型能力增强，智能体框架应随时间缩小。
Agent harnesses should shrink over time as models become more capable.
简单的文件系统用于记忆往往优于复杂的索引存储。
Simple file systems for memory often outperform complex indexed stores.
智能体应拥有自己的权限和身份，而非继承用户的。
Agents should have their own permissions and identities, not assume the user's.
营销等非工程团队也能有效采用智能体工具。
Non-engineering teams like marketing can adopt agentic tools effectively.

本期章节 · Chapters（共 22）

Claude Tag简介 Introduction to Claude Tag
什么是Claude Tag？ What is Claude Tag?
Slack中@Claude与Claude Tag的区别 Difference between @Claude in Slack and Claude Tag
内部使用示例与上下文记忆 Internal usage example and context memory
Claude Tag的上下文与记忆 Claude Tag's Context and Memory
Slack中的端到端工作流 End-to-End Workflow in Slack
Claude Tag作为编排器 Claude Tag as Orchestrator
工作流变革：从单人游戏到多人游戏 Workflow Change: From Single-Player to Multiplayer
Claude Tag vs Claude Code：异步与同步 Claude Tag vs Claude Code: Asynchronous vs Synchronous
模型自我验证工作 Models verifying their own work
Slack作为智能体界面 Slack as a surface for agents
聊天与IDE中定义“好”的标准 Defining what good looks like in chat vs IDE
Claude Code的外部采用与规模 External adoption and scale of Claude Code
自建与购买及人类适应速度 Build vs buy and human adaptation speed
AI开发速度与基础设施挑战 Pace of AI development and infrastructure challenges
编码之外的智能体工具 Agentic tools beyond coding
为未来模型构建并简化框架 Building for future models and simplifying harnesses
Claude在销售与产品开发中的主动与定时任务 Claude's proactive and scheduled tasks in sales and product development
事件响应智能体设计原则与信任建立 Incident response agent design principles and trust journey
托管智能体中的“梦想”功能 Dreaming feature in managed agents
每日简报与团队报告 Daily Brief and Team Reports
结束语 Closing Remarks

阅读全文双语转录 →