Daily Research Brief 2026-08-07

📊 Token usage: estimated from retrieval and writing scale.

Covers the latest AI research, open source and industry moves, updated daily.

Editor’s Note

Papers and industry collide on the same point today: long-horizon reliability no longer comes from swapping models but from the ‘shell’. OneDayAgent hits 0.821 across five backend models with one harness, Mimir separates world memory from task memory for a 42.5% peak gain, and LeanMem sorts memory by compressibility.

1. Latest arXiv Papers

1. OneDayAgent: Towards a Long-Horizon Harness for Autonomous Agents — https://arxiv.org/abs/2608.05013

2. Mimir: A Neuro-Symbolic Memory System with Dynamic Grounding for Embodied Agents — https://arxiv.org/abs/2608.04933

3. LeanMem: Simple and Efficient Long-Term Memory for LLM Agents — https://arxiv.org/abs/2608.03463

4. Toward Skill-Native LLMs: Skill Entropy for Benchmarking and Training Long-Horizon Reasoning — https://arxiv.org/abs/2608.05139

5. Unified Agent: Managing Interactions across Devices — https://arxiv.org/abs/2608.05729

6. Agentic Reinforcement Learning with Observation-Calibrated Self-Distillation (OCSD) — https://arxiv.org/abs/2608.04788

7. TurnSight: Turn-Level Hindsight Self-Distillation for Tool-Integrated Reasoning — https://arxiv.org/abs/2608.04007

8. Stress-Testing AI Agents in a Real Machine-Catalysis Laboratory (USTC) — https://arxiv.org/abs/2607.23045