Search Ctrl+K

Change language Switch ThemeSign In

Curated Daily BriefWeekly PicksTopics SettingsHelp CenterCollapse

Narrow Mode

CodexLoom: Turning Codex Threads into Long-Term Digital Employees Organized as an Agent Team

CodexLoom: Turning Codex Threads into Long-Term Digital Employees Organized as an Agent Team

@aigclink

前manus大佬开源的agent team产品codexloom：把 Codex 的 thread 变成长期在岗的数字员工，再组织成一支团队。解决了一个人无法同时高效管理几十个agent的场景，一旦agent多了人就是最大的瓶颈，主要面向手里养着数十以上专职 agent、想让agent各守一个细分场景、还能被同事或用户通过飞书直接调用的高级个人/一人公司。

现在大家用 code agent 都是任务式的：开 thread，交代背景，拿结果，关掉。旧 thread 里积累的约束、决策、工具用法全部作废，每个任务都在重复付冷启动成本。

CodexLoom 反着做：旧的thread 不该随任务废弃，同场景的后续工作回到同一条 thread。

有人担心长 thread 太贵，作者的回应很扎实：

1、新开 thread 重建上下文同样有成本，只是没人算这笔账

2、稳定的历史前缀能吃 prompt cache

3、compaction 压缩旧消息后，工作状态依然在

4、最关键：纠错、术语、偏好、判断习惯这些隐性知识，换 thread 就丢，用 prompt 根本重建不出来

每个 Agent 有个profile，包含三样东西：Identity（是谁）、Domain（长期负责什么）、Scope（边界在哪）。一份数字员工的岗位说明书。

通信规则：信息流向责任，决定流向权限，结果回到发起者，例外向上升级。拿去管人类团队同样成立。

最有想象空间的是对外：养成熟的 Agent 带着受治理的身份进飞书/Slack 群，同事直接跟它协作。你调教半年的能力，别人可以直接复用。一人公司雇数字员工再外派，最小闭环跑通了，也是未来AI数字员工商业化的基础。

这个项目的方法论非常值得学习，别人还在比谁的编排 DAG 花哨，它在回答更根本的问题：Agent 从一次性工具变成长期同事，组织关系该长什么样。工龄，可能才是下一代 Agent 产品真正的护城河。

#CodexLoom #Codex #AIagent #多智能体 #agent治理 Show More

!Image 3: Tweet image

Aug 2, 2026, 1:39 AM View on X

11 Replies

15 Retweets

71 Likes

6,478 Views  AIGCLINK @aigclink

Follow

One Sentence Summary

The ex-Manus builder open-sources CodexLoom, turning Codex threads into long-running agent workflows with profiles, scope permissions, and Feishu/Slack collaboration.

Summary

The tweet provides an in-depth analysis of CodexLoom: it turns Codex threads into long-term “digital employees” that stay on the job, then organizes them into a team — solving the bottleneck of one person trying to manage dozens of Agents. Each Agent has a profile with three elements — Identity, Domain, and Scope — acting as a job description. Follow-up work in the same scenario returns to the same thread, avoiding repeated cold-start costs. The author also responds to concerns about long-thread costs: rebuilding context in a new thread also costs, stable history prefixes benefit from prompt caching, compaction preserves working state, and tacit knowledge (corrections, terminology, preferences, judgment habits) is lost when switching threads. The project supports mature Agents entering Feishu/Slack groups with governed identities for direct collaboration, laying the foundation for the future commercialization of AI digital employees.

AI Screening

87

Influence Score 45

Published Today

Language

Chinese

Tags

CodexLoom

AI Agent

Multi-Agent

Agent Governance

Codex

Make your daily reading actually fit you.A daily brief built from the sources you follow. Get started free HomeDiscoverSettings