{
  "schema_version": "1.1",
  "id": "s8:https://www.youtube.com/watch?v=2bvtay8wGYI",
  "slug": "scaling-to-long-horizons-ross-taylor-chengxi-taylor-general-reasoning-0jwtg4d",
  "url": "https://feed7.dev/p/scaling-to-long-horizons-ross-taylor-chengxi-taylor-general-reasoning-0jwtg4d",
  "title": "Scaling to Long Horizons — Ross Taylor & Chengxi Taylor, General Reasoning",
  "why_included": "Long-running agents strain context, reward design, and GPU scheduling at once. Compaction and value models can help, but each introduces another optimization trade-off.",
  "summary": "Long-horizon RL faces gradient variance that grows with trajectory length, sparse terminal rewards, credit assignment, and variable episode lengths. A **1 million-token context window** is still small relative to the horizons the speakers target.",
  "practical_implication": "Plan agent state beyond the prompt: use compaction, file-based scratch space, trajectory search, or archives as the task permits. For training, critics can provide earlier signals, while pipeline RL trades fresher policy data for better GPU use.",
  "agent_context": "Long-horizon RL faces gradient variance that grows with trajectory length, sparse terminal rewards, credit assignment, and variable episode lengths. A **1 million-token context window** is still small relative to the horizons the speakers target.\n\nPlan agent state beyond the prompt: use compaction, file-based scratch space, trajectory search, or archives as the task permits. For training, critics can provide earlier signals, while pipeline RL trades fresher policy data for better GPU use.\n\nThe team reports pipeline training tolerating about **eight off-policy steps**, but week-long inference can exceed that window. Bootstrapping with a value model keeps hardware occupied at the cost of bias, and its platform’s **350 environments** do not remove that trade-off.",
  "source": {
    "name": "AI Engineer",
    "url": "https://www.youtube.com/watch?v=2bvtay8wGYI",
    "published_at": "2026-07-31T21:30:06.000Z"
  },
  "source_class": "video",
  "content_type": "Video",
  "layer": "agent",
  "domains": [
    "coding",
    "data"
  ],
  "topics": [
    "harness-engineering",
    "agent-reliability",
    "context-engineering"
  ],
  "verification": {
    "status": "source_linked",
    "label": "Source Linked",
    "method": "source_feed",
    "verified_at": null
  },
  "uncertainty": [
    "The team reports pipeline training tolerating about **eight off-policy steps**, but week-long inference can exceed that window. Bootstrapping with a value model keeps hardware occupied at the cost of bias, and its platform’s **350 environments** do not remove that trade-off."
  ],
  "connected_context": {
    "meaning": "This establishes long-horizon agency as a state-management and training-systems problem that a larger prompt alone does not solve. It deepens prior harness guidance by requiring durable state outside context and exposes a deployment gap: pipeline RL may tolerate limited policy staleness, while week-long trajectories can exceed it, leaving a trade-off among hardware use, fresh data, critic bias, and sparse rewards.",
    "corpus_size": 318,
    "generated_at": "2026-08-01T10:08:44.175Z",
    "connections": [
      {
        "title": "Data and Environment Curation for Post-Training LLMs — Mahesh Sathiamoorthy, Bespoke Labs",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=ewtOo0scUh0",
        "feed7_url": "https://feed7.dev/p/data-and-environment-curation-for-post-training-llms-mahesh-sathiamoorth-19ab77a",
        "reason": "Bespoke’s SFT-first guidance becomes more consequential here because long-horizon RL adds sparse rewards, growing gradient variance, policy staleness, and infrastructure costs beyond ordinary curation."
      },
      {
        "title": "Do AI Agents Know When a Task Is Simple? Toward Complexity-Aware Reasoning and Execution",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2607.13034v1",
        "feed7_url": "https://feed7.dev/p/2607-13034v1-03g7ghx",
        "reason": "E3’s minimum-viable-path strategy can avoid unnecessarily long trajectories, whereas this signal addresses the durable state and training problems that remain when the task genuinely requires a long horizon."
      },
      {
        "title": "AI tools for Forward Deployed Engineering — Vasuman Moza, Varick Agents",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=l0FLhNqBOic",
        "feed7_url": "https://feed7.dev/p/ai-tools-for-forward-deployed-engineering-vasuman-moza-varick-agents-12kjg79",
        "reason": "Varick’s explicit mapping of ownership, exceptions, and handoffs provides the process structure that long-running agents would need to preserve through compaction, archives, or file-based state."
      },
      {
        "title": "Let's integrate AI Agents in Event-Sourced Systems — Divakar Kumar, FlyersSoft",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=o6U_2vd967Y",
        "feed7_url": "https://feed7.dev/p/let-s-integrate-ai-agents-in-event-sourced-systems-divakar-kumar-flyerss-1w3t01t",
        "reason": "The event-sourced design offers a concrete external-state pattern for bounded context and resumable decisions, complementing this signal’s warning that prompt context cannot hold the full operational horizon."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-07-31T21:30:06.000Z",
  "modified_at": "2026-07-31T21:30:06.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/scaling-to-long-horizons-ross-taylor-chengxi-taylor-general-reasoning-0jwtg4d",
    "json": "https://feed7.dev/p/scaling-to-long-horizons-ross-taylor-chengxi-taylor-general-reasoning-0jwtg4d.json",
    "markdown": "https://feed7.dev/p/scaling-to-long-horizons-ross-taylor-chengxi-taylor-general-reasoning-0jwtg4d.md"
  }
}