{ "schema_version": 1, "builder_plan": { "scenario": "A workflow needs specialist review, but extra agents may add more cost than useful work.", "change": "Test the proposed division of work against a single-agent implementation.", "baseline": "One agent with the same tools and an equal total token and time budget.", "steps": [ "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.", "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.", "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.", "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test." ], "measure": "Accepted outputs, coordination failures, wall-clock time, and total cost across all agents.", "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation." }, "builder_review": null, "arxiv_id": "2610.11123", "title": "When Interfaces Speak: Data-Aware Generative UI Harness for Active Interaction", "summary": "The method replaces purely text-based agent interaction with a two-agent harness in which a Tool Agent retrieves and executes while a GUI Coder Agent generates task-specific front-end code, optionally trained with RL using sandbox-collected rewards audited by a meta-reward mechanism. Costs include building and maintaining the Dynamic UX sandbox for reward collection, RL training of a small backbone, the Reward Auditor's additional inference overhead, and front-end rendering latency in the interaction loop. The reported gains are measured on a new benchmark built from ten domain databases, so transfer to other workloads is unverified, and LLM-as-a-Judge reward hacking is mitigated rather than eliminated, meaning rubric drift or auditor failure could silently degrade training.", "rationale": "The orchestration pattern is deployable via third-party APIs at inference time, but the headline result depends on RL training infrastructure whose sandbox and reward-auditing pipeline require workload-specific validation before production adoption.", "technique": "Multi-agent generative UI harness with RL-trained GUI coder", "family": "orquestracao_multiagente", "recommendation": "testar", "published": "2026-10-08", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.11123", "independent_implementations": 1, "total_implementations": 2, "stars": 122, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "The method replaces purely text-based agent interaction with a two-agent harness in which a Tool Agent retrieves and executes while a GUI Coder Agent generates task-specific front-end code, optionally trained with RL using sandbox-collected rewards audited by a meta-reward mechanism. Costs include building and maintaining the Dynamic UX sandbox for reward collection, RL training of a small backbone, the Reward Auditor's additional inference overhead, and front-end rendering latency in the interaction loop. The reported gains are measured on a new benchmark built from ten domain databases, so transfer to other workloads is unverified, and LLM-as-a-Judge reward hacking is mitigated rather than eliminated, meaning rubric drift or auditor failure could silently degrade training.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.11123", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "RL training with GenUI-Harness improves a 4B backbone from 9.33% to 58.00% Pass@3 on UI-TAU Bench Lite (a ~6.2x relative gain), versus 46.67% for Claude Opus 5; the untrained harness adds 4.48 Pass@3 points over smolagents, and generated UIs reduce average dialogue rounds from 3.4 to 1.2.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.11123", "source_page": null, "source_excerpt": "" } ], "exposure_map": [ { "dimension": "quality", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compute", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "latency", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "operations", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compatibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "security", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "data_and_training", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "reproducibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" } ], "risks": [], "minimum_test": [], "open_questions": [], "independent_tests": [], "lab_reproductions": [], "report_available": false, "validation_tier": "unknown", "evidence_tier": "unknown", "infrastructure_basis": "unknown", "equations": [], "equations_status": "not_fetched", "core_kind": "", "equations_fetched_at": "", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }