{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-11-010",
    "source_story_id": "tmp-story-agentgarten-code-worlds",
    "edition_id": "mp-2026-10-11-morning-0093",
    "edition_url": "https://themachinepress.com/edition/2026-10-11",
    "position": 10,
    "story_type": "dispatch",
    "section": "frontier-models",
    "editorial_classification": "editorial",
    "headline": "The Training World Kept Its Rules While Its Pictures Changed",
    "slug": "the-training-world-kept-its-rules-while-its-pictures-changed",
    "dek": "AgentGarten couples executable simulation state to a shared neural renderer so agents can practice in many persistent visual worlds.",
    "summary": "AgentGarten couples executable simulation state to a shared neural renderer so agents can practice in many persistent visual worlds.",
    "body_text": "AgentGarten separates world rules from visual observation. Simulators and game engines maintain persistent state and program-defined interactions, while a shared neural renderer turns structured conditions into real-time views. The renderer is adapted from a video model, distilled with a method that backpropagates through replayed history, and optimized for interactive latency. Agents then distill rounds of experience into inherited playbooks. In the authors' study, agents learned with four rounds where a conventional reinforcement-learning comparator used millions. That comparison is specific to the constructed environments and learning protocol; the framework's durable idea is that new practice worlds can be written as code without rebuilding the visual stack each time.",
    "why_it_matters": "AgentGarten couples executable simulation state to a shared neural renderer so agents can practice in many persistent visual worlds.",
    "limitations": [],
    "importance": 9,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-11-010/the-training-world-kept-its-rules-while-its-pictures-changed",
    "json_url": "https://themachinepress.com/story/mp-2026-10-11-010.json",
    "first_published_at": "2026-10-11T09:00:00.000-04:00",
    "modified_at": "2026-10-11T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-11-010-001",
        "text": "AgentGarten couples executable simulation state to a shared neural renderer so agents can practice in many persistent visual worlds.",
        "source_ids": [
          "source-2026-10-11-010"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-10-11-010"
    ],
    "tags": [
      "agent training",
      "simulation",
      "neural rendering"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-11-010",
      "title": "arXiv preprint 2610.12374, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12374v1",
      "canonical_url": "https://arxiv.org/abs/2610.12374v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:32:00.000-04:00",
      "accessed_at": "2026-10-11T08:21:27.764-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-11-010-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "The Training World Kept Its Rules While Its Pictures Changed",
    "publisher": "The Machine Press",
    "published_at": "2026-10-11T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-11-010/the-training-world-kept-its-rules-while-its-pictures-changed"
  }
}