{
  "schema_version": "1.1",
  "id": "s13:https://arxiv.org/abs/2609.10522v1",
  "slug": "2609-10522v1-1lobtfw",
  "url": "https://feed7.dev/p/2609-10522v1-1lobtfw",
  "title": "Show-Harness: Just a VLM Agent Can Play Robots",
  "why_included": "Show-Harness puts a compact semantic action layer between VLM intent and robot-specific control, a harness pattern for reusing frontier or small open models across embodiments.",
  "summary": "Show-Harness exposes **discrete semantic action units** that a VLM selects, then uses embodiment-specific interpreters to ground them deterministically into local robot actions. **GUMI** applies the same action space to GUI-based demonstrations without specialized teleoperation hardware.",
  "practical_implication": "The transferable lesson for agent builders is to place a small, model-readable interface above deterministic adapters. The paper reports zero-shot use of closed frontier VLMs and adaptation of small open VLMs with **a few GPU-hours** of fine-tuning.",
  "agent_context": "Show-Harness exposes **discrete semantic action units** that a VLM selects, then uses embodiment-specific interpreters to ground them deterministically into local robot actions. **GUMI** applies the same action space to GUI-based demonstrations without specialized teleoperation hardware.\n\nThe transferable lesson for agent builders is to place a small, model-readable interface above deterministic adapters. The paper reports zero-shot use of closed frontier VLMs and adaptation of small open VLMs with **a few GPU-hours** of fine-tuning.\n\nThe supplied abstract claims gains across tasks, robots, and environments, but provides no numerical results or deployment failure analysis. The approach still leaves fine-grained physical decisions with the VLM, so real-world safety remains an open concern.",
  "source": {
    "name": "arXiv",
    "url": "https://arxiv.org/abs/2609.10522v1",
    "published_at": "2026-09-09T17:53:38.000Z"
  },
  "source_class": "blog_post",
  "content_type": "Paper",
  "layer": "agent",
  "domains": [
    "image"
  ],
  "topics": [
    "harness-engineering",
    "tool-use",
    "open-models"
  ],
  "verification": {
    "status": "needs_review",
    "label": "Needs Review",
    "method": "unverified",
    "verified_at": null
  },
  "uncertainty": [
    "The supplied abstract claims gains across tasks, robots, and environments, but provides no numerical results or deployment failure analysis. The approach still leaves fine-grained physical decisions with the VLM, so real-world safety remains an open concern."
  ],
  "connected_context": {
    "meaning": "This confirms a reusable harness pattern beyond software: expose a compact semantic action vocabulary to the model and keep embodiment-specific execution in deterministic adapters. GUMI broadens how demonstrations can be collected, while the missing numerical and failure evidence prevents treating interface simplification as proof of safe or reliable physical autonomy.",
    "corpus_size": 732,
    "generated_at": "2026-09-10T10:09:53.895Z",
    "connections": [
      {
        "title": "Tell the Robot What You Want — Sandhya Subramani, AWS",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=S6aSoQ6_u5A",
        "feed7_url": "https://feed7.dev/p/tell-the-robot-what-you-want-sandhya-subramani-aws-1i532jq",
        "reason": "Both put a model above existing robot capabilities; Show-Harness formalizes the shared semantic action layer and deterministic embodiment adapters, while the AWS demo’s failures show why execution and completion still need verification."
      },
      {
        "title": "Multi-Step Tool-Calling over Korean Open Public APIs: A Benchmark and a Data-Synthesis Recipe",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2609.05395v1",
        "feed7_url": "https://feed7.dev/p/2609-05395v1-0qwa6p2",
        "reason": "EDGE applies a parallel principle to APIs: constrain model tool use through structured actions and deterministic live validation. Together they support separating model-readable choices from environment-specific execution."
      },
      {
        "title": "Give the Agent a Budget, Not a Token — Sachin Malhotra, Anthropic",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=rbjWzZK2LU0",
        "feed7_url": "https://feed7.dev/p/give-the-agent-a-budget-not-a-token-sachin-malhotra-anthropic-1sm80qu",
        "reason": "Deterministic action grounding does not bound an agent’s overall physical authority; runtime limits, trip wires, and human-held overrides remain necessary when selected actions can cause consequential changes."
      },
      {
        "title": "AI Agents Are Just Distributed Systems Now — Salman Munaf, TikTok",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=hD9-V56FNRI",
        "feed7_url": "https://feed7.dev/p/ai-agents-are-just-distributed-systems-now-salman-munaf-tiktok-1v4yc47",
        "reason": "The adapters make execution deterministic locally, but external actions can still have ambiguous outcomes, so idempotency, durable traces, reconciliation, and recovery remain harness requirements."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-09-09T17:53:38.000Z",
  "modified_at": "2026-09-09T17:53:38.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/2609-10522v1-1lobtfw",
    "json": "https://feed7.dev/p/2609-10522v1-1lobtfw.json",
    "markdown": "https://feed7.dev/p/2609-10522v1-1lobtfw.md"
  }
}