arxiv arXiv cs.CL · 1h ago · research

Living-Harness self-evolves agent harnesses to improve Pass@1 by ~10 points

from English

Researchers propose Living-Harness, a self-evolving agent harness that converts completed trajectories and evaluator signals into posterior evidence for bounded harness updates. Guided by an Evolution-SOP, the system extracts episode abstractions and structured update evidence to write episodic memory and state graphs.

Living-Harness records trigger conditions, failure patterns, and recovery actions as episodic memory.
It maintains a state graph of nodes, repair edges, and transition rules for procedural repairs.
The updated harness state is retrieved to guide future interactions while tools and base context remain frozen.
On eight interactive environments derived from τ²-Bench and MultiWOZ-2.4, it improves average Pass@1 over the strongest baseline by 10.07 and 9.91 percentage points respectively.

The evolved harness state supports retrieval-only reuse across model backbones, allowing procedural repairs to accumulate across evolution cycles.

Importance 2/3 New harness with differentiators arXiv cs.CL AI agents Research paper

Read original

media Hugging Face Forums · 2d ago

Three 2026 papers argue distributed agent ecosystems outperform monolithic models

Three recent academic papers from Google, Harvard & MIT, and Imperial College & Huawei converge on the conclusion that the next AI intelligence explosion will be driven by distributed systems rather than larger single models. The authors argue that intelligence is fundamentally plural, emerging from the interaction of specialized agents or internal societal structures.

arxiv arXiv cs.CL · 2d ago

AgentOmnia scales agentic models across full-scenario applications

Researchers present AgentOmnia, a framework for full-scenario agentic scaling that coordinates task-space definition, data synthesis, post-training, evaluation, and improvement across To-Consumer, To-Business, and To-Employee applications. The system utilizes an extensible taxonomy and constructs 5,018 stateful environments with 255,375 tools and 52,361 tasks to enable fine-grained diagnosis via OmniaBench.

lab NVIDIA Research · 2d ago · 1 view

NVIDIA researchers propose field experiments in Rocket League tournaments

Researchers at NVIDIA are addressing the difficulty of scheduling competitive group play and recreating lab environments by adopting field experiment methodology. This approach integrates experimentation directly into tournament structures, creating hybrid tournament-experiments.

lab NVIDIA Research · 6d ago

AITE framework autonomously discovers wireless algorithms via LLM-driven search

Researchers introduce The AI Telco Engineer (AITE), a framework that uses large language model-driven evolutionary search to autonomously design algorithms for complex communication problems while navigating performance-complexity tradeoffs.

arxiv arXiv cs.AI · 6d ago · 3 views

AREX introduces recursively self-improving deep research agents

Researchers introduce AREX, a family of Recursively Self-Improving (RSI) deep research agents designed to address the discovery-verification asymmetry in complex queries. AREX alternates between an inner loop that gathers evidence and constructs provisional answers, and an outer loop that audits constraints to guide targeted refinement.