{
  "version": 1,
  "event_id": "evt_eb56a02846741ca3",
  "url": "https://xiyu.news/events/evt_eb56a02846741ca3/",
  "json": "https://xiyu.news/api/events/evt_eb56a02846741ca3.json",
  "type": "other",
  "status": "monitoring",
  "category": "technology",
  "title": {
    "zh": "Anthropic披露Claude Code团队以AI审查AI的大规模代码审查实践",
    "en": "Anthropic details AI-reviewing-AI code review practice for Claude Code team"
  },
  "current_state": {
    "zh": "Anthropic 披露，Claude Code 约 90% 的代码如今由 Claude 编写，并已围绕多智能体 AI 检查与按风险分级的人工审批重建代码审查流程，同时对失控的审查循环设限——此前有一次改动在一夜之间经历了 50 多轮审查。",
    "en": "Anthropic disclosed that roughly 90% of Claude Code's code is now written by Claude and that it has rebuilt code review around multi-agent AI checking with risk-tiered human approval, while capping runaway review loops after one change went through 50+ rounds in a single night."
  },
  "first_seen_at": "2026-10-16T23:27:32.709086+00:00",
  "last_updated_at": "2026-10-18T13:01:27.109904+00:00",
  "last_material_change_at": "2026-10-18T13:01:27.109904+00:00",
  "confidence": 0.75,
  "updates_count": 2,
  "sources_count": 2,
  "entities": [
    "claude",
    "claude-code",
    "code",
    "code-review",
    "how"
  ],
  "identifiers": [],
  "topics": [
    "agentic-coding",
    "ai-code-review",
    "ai-devtools",
    "anthropic",
    "claude-code",
    "code-review",
    "engineering-practices",
    "software-engineering"
  ],
  "updates": [
    {
      "update_id": "upd_df6d56bb6aa2b516",
      "event_id": "evt_eb56a02846741ca3",
      "occurred_at": "2026-10-16T19:04:47Z",
      "published_at": "2026-10-16T19:04:47Z",
      "first_seen_at": "2026-10-16T23:27:32.709086Z",
      "time_precision": "published",
      "update_type": "initial",
      "material_change": true,
      "title_zh": "Claude Code 团队如何做代码审查",
      "title_en": "How the Claude Code team does code review",
      "what_changed_zh": "Anthropic 的 Claude Code 团队发布博文，介绍其大规模代码审查流程：据其描述每天处理约 3 万个 PR，大部分路径已自动化，人的注意力被引导到高风险改动上。文档、文案、仅内部使用的功能以及带开关的增量改动等低风险改动，只要自动审查与安全检查通过，就不需要人工批准。\n\n按照该团队自己的表述，这套流程的目的是把有限的人工审查精力只放到有风险的地方；同时评估套件（eval suite）会随每一次事故扩充，审查器的每一次改动都要针对整个套件评分。\n\n每天约 3 万个 PR 的数字以及低风险类别（文档、文案、仅内部功能、带开关的增量改动）均出自该团队自己的博文，而非独立测算。",
      "what_changed_en": "Anthropic's Claude Code team published a post describing how it runs code review at scale, reporting roughly 30,000 pull requests per day, with most of the path automated and human attention directed to high-risk changes. Low-risk changes — docs, copy, internal-only features and additive changes behind a flag — require no human approval once automated review and security checks are green.\n\nAccording to the team's own framing, the pipeline is designed to send scarce human review attention only where the risk is, while the reviewer's eval suite grows with every incident and every change to the reviewer is scored against all of it.\n\nThe ~30k PRs/day figure and the low-risk categories (docs, copy, internal-only features, additive changes behind a flag) come from the team's own blog post rather than independent measurement.",
      "current_state_zh": "Anthropic 的 Claude Code 团队发布博文，介绍其大规模代码审查流程：据其描述每天处理约 3 万个 PR，大部分路径已自动化，人的注意力被引导到高风险改动上。文档、文案、仅内部使用的功能以及带开关的增量改动等低风险改动，只要自动审查与安全检查通过，就不需要人工批准。\n\n按照该团队自己的表述，这套流程的目的是把有限的人工审查精力只放到有风险的地方；同时评估套件（eval suite）会随每一次事故扩充，审查器的每一次改动都要针对整个套件评分。\n\n每天约 3 万个 PR 的数字以及低风险类别（文档、文案、仅内部功能、带开关的增量改动）均出自该团队自己的博文，而非独立测算。",
      "current_state_en": "Anthropic's Claude Code team published a post describing how it runs code review at scale, reporting roughly 30,000 pull requests per day, with most of the path automated and human attention directed to high-risk changes. Low-risk changes — docs, copy, internal-only features and additive changes behind a flag — require no human approval once automated review and security checks are green.\n\nAccording to the team's own framing, the pipeline is designed to send scarce human review attention only where the risk is, while the reviewer's eval suite grows with every incident and every change to the reviewer is scored against all of it.\n\nThe ~30k PRs/day figure and the low-risk categories (docs, copy, internal-only features, additive changes behind a flag) come from the team's own blog post rather than independent measurement.",
      "detailed_summary_zh": "Anthropic 的 Claude Code 团队发布博文，介绍其大规模代码审查流程：据其描述每天处理约 3 万个 PR，大部分路径已自动化，人的注意力被引导到高风险改动上。文档、文案、仅内部使用的功能以及带开关的增量改动等低风险改动，只要自动审查与安全检查通过，就不需要人工批准。\n\n按照该团队自己的表述，这套流程的目的是把有限的人工审查精力只放到有风险的地方；同时评估套件（eval suite）会随每一次事故扩充，审查器的每一次改动都要针对整个套件评分。\n\n每天约 3 万个 PR 的数字以及低风险类别（文档、文案、仅内部功能、带开关的增量改动）均出自该团队自己的博文，而非独立测算。",
      "detailed_summary_en": "Anthropic's Claude Code team published a post describing how it runs code review at scale, reporting roughly 30,000 pull requests per day, with most of the path automated and human attention directed to high-risk changes. Low-risk changes — docs, copy, internal-only features and additive changes behind a flag — require no human approval once automated review and security checks are green.\n\nAccording to the team's own framing, the pipeline is designed to send scarce human review attention only where the risk is, while the reviewer's eval suite grows with every incident and every change to the reviewer is scored against all of it.\n\nThe ~30k PRs/day figure and the low-risk categories (docs, copy, internal-only features, additive changes behind a flag) come from the team's own blog post rather than independent measurement.",
      "background_zh": "该博文出自 Anthropic 的 Claude Code 团队；Claude Code 是 Anthropic 的智能体式编程工具，最初以限定研究预览形式发布，并通过 npm 与 winget 分发。自动代码审查本身是一个成熟的工具类别，指在不经人工干预的情况下用软件分析源代码中的安全漏洞、缺陷与编码规范违规。",
      "background_en": "The post comes from Anthropic's Claude Code team; Claude Code is Anthropic's agentic coding tool, launched in a limited research preview and distributed via npm and winget. Automated code review is itself an established tooling category that analyzes source code for security vulnerabilities, bugs and coding-standard violations without human intervention.",
      "community_discussion_zh": "在 Hacker News 的讨论（189 分、179 条评论）中，整体氛围偏批评而非宣传：有人质疑如此规模下的代码流失率与每行代码成本，有人怀疑 AI 审查 AI 生成的代码、每天 3 万个 PR 时“审查”还能否名副其实，也有人认为文风过于精致、未必反映真实做法；还有评论者反对让大模型写文档却无人阅读。",
      "community_discussion_en": "The Hacker News thread (189 points, 179 comments) was critical rather than promotional: commenters questioned the churn rate and cost per line of code at that volume, whether AI reviewing AI-generated code at 30k PRs a day can meaningfully be called review, and whether the polished prose reflects genuine practice; one objected to letting an LLM write docs that nobody reads.",
      "market_impact_zh": "",
      "market_impact_en": "",
      "importance_score": 7.0,
      "references": [
        {
          "url": "https://claude.com/product/claude-code",
          "title": "Claude Code by Anthropic | AI Coding Agent, Terminal, IDE"
        },
        {
          "url": "https://github.com/anthropics/claude-code",
          "title": "GitHub - anthropics/claude-code: Claude Code is an agentic coding..."
        },
        {
          "url": "https://www.wiz.io/academy/application-security/automated-code-review",
          "title": "What is automated code review? Tools and best practices | Wiz"
        },
        {
          "url": "https://news.ycombinator.com/item?id=50122951",
          "title": "Community discussion"
        }
      ],
      "confidence": 0.75,
      "story_ids": [
        "hackernews:story:50122951"
      ],
      "sources": [
        {
          "url": "https://claude.dev/blog/solving-code-review-at-anthropic/",
          "label": "pretext",
          "source_type": "hackernews",
          "official": false
        }
      ]
    },
    {
      "update_id": "upd_41e6200caf8f1079",
      "event_id": "evt_eb56a02846741ca3",
      "occurred_at": "2026-10-18T11:12:02Z",
      "published_at": "2026-10-18T11:12:02Z",
      "first_seen_at": "2026-10-18T13:01:27.109904Z",
      "time_precision": "published",
      "update_type": "confirmation",
      "material_change": true,
      "title_zh": "Claude Code开始让AI审查AI：90%代码由Claude编写，低风险修改无需人工批准",
      "title_en": "Claude Code开始让AI审查AI：90%代码由Claude编写，低风险修改无需人工批准",
      "what_changed_zh": "Anthropic 披露，Claude Code 约 90% 的代码如今由 Claude 编写，并已围绕多智能体 AI 检查与按风险分级的人工审批重建代码审查流程，同时对失控的审查循环设限——此前有一次改动在一夜之间经历了 50 多轮审查。",
      "what_changed_en": "Anthropic disclosed that roughly 90% of Claude Code's code is now written by Claude and that it has rebuilt code review around multi-agent AI checking with risk-tiered human approval, while capping runaway review loops after one change went through 50+ rounds in a single night.",
      "current_state_zh": "Anthropic 披露，Claude Code 约 90% 的代码如今由 Claude 编写，并已围绕多智能体 AI 检查与按风险分级的人工审批重建代码审查流程，同时对失控的审查循环设限——此前有一次改动在一夜之间经历了 50 多轮审查。",
      "current_state_en": "Anthropic disclosed that roughly 90% of Claude Code's code is now written by Claude and that it has rebuilt code review around multi-agent AI checking with risk-tiered human approval, while capping runaway review loops after one change went through 50+ rounds in a single night.",
      "detailed_summary_zh": "Anthropic disclosed that roughly 90% of Claude Code's code is now written by Claude and that it has rebuilt code review around multi-agent AI checking with risk-tiered human approval, while capping runaway review loops after one change went through 50+ rounds in a single night.",
      "detailed_summary_en": "Anthropic disclosed that roughly 90% of Claude Code's code is now written by Claude and that it has rebuilt code review around multi-agent AI checking with risk-tiered human approval, while capping runaway review loops after one change went through 50+ rounds in a single night.",
      "background_zh": "",
      "background_en": "",
      "community_discussion_zh": "",
      "community_discussion_en": "",
      "market_impact_zh": "",
      "market_impact_en": "",
      "importance_score": 7.0,
      "references": [],
      "confidence": 0.75,
      "story_ids": [
        "telegram:theblockbeats:201087"
      ],
      "sources": [
        {
          "url": "https://m.theblockbeats.info/flash/372426?from=telegram",
          "label": "theblockbeats",
          "source_type": "telegram",
          "official": false
        }
      ]
    }
  ]
}