{
  "id": 14395873,
  "title": "After shutting down the control surface, the AI task is still running: Why did we break it down into three layers?",
  "url": "https://urgent.news/2026/10/14/ai-14395873",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-14T07:49:27.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/edgestorage/ting-diao-kong-zhi-mian-zhi-hou-ai-ren-wu-huan-zai-pao-wo-men-wei-shi-yao-ba-ta-chai-cheng-san-ceng-4hdp"
  },
  "original_language": "zh",
  "account": "TaskHandoff has developed an architecture to address issues that arise when using AI to write code across multiple machines. The system consists of three layers: Control Plane, Node Agent, and Controlled Instance, each with distinct responsibilities and state ownership. This design allows tasks to continue running even if the control panel is closed or restarted, and tasks can be monitored and managed across multiple machines and platforms. The system's design is based on four key principles, including separating state ownership and distinguishing between display status and authoritative status. Documentation is available for users to deploy and use the system.",
  "summary": "Interacting with a terminal window on a machine through an interface is how most people start using AI to write code. As the number of tasks increases, three specific problems arise: tasks are scattered across multiple machines, each task requires an independent and disposable environment, and long-running tasks should not die when the browser is closed or the interface is disconnected. This article discusses the architectural choices made by TaskHandoff to address these three issues: why the control plane and execution plane are separated, who manages the state, and the resulting trade-offs. The problem: when tasks multiply, you start manually acting as a scheduler Suppose you have three machines: a laptop, a constantly running home server, and a company intranet host. You run Codex or OpenCode on each of them. As a result, you spend your days doing the following: remembering which machine has which repository, which branch, and which worktree; manually setting up the environment or tolerating the container's slow installation of tools from scratch; repeatedly switching SSH windows to check task progress, as tasks only exist within the context of that terminal; and after closing your laptop, wondering if the task on the home server is still running.",
  "key_points": [
    "TaskHandoff separates control plane from execution plane to address AI task management issues.",
    "Control plane manages node lists, instance dashboards, AI sessions, and credentials."
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 3,
    "also_reported_by": [
      {
        "outlet": "Hankyoreh",
        "title": "Japanese new lawyers 'I'm shaking... Global 'legal AI' lands one after another'",
        "url": "https://urgent.news/2026/10/14/ai-14391751",
        "published": "2026-10-14T05:14:55.000Z"
      },
      {
        "outlet": "Asahi Business",
        "title": "AI takes control of smart home appliances and proliferates? Anthropic warns of research results",
        "url": "https://urgent.news/2026/10/14/ai-14405411",
        "published": "2026-10-14T08:30:35.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}