Daily Research Brief 2026-07-30

📊 Token usage: estimated from retrieval and writing scale.

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

Editor’s Note

The signal to watch today: ‘agents are evolving from tools into self-evolving systems’. On arXiv, SkillRise makes cross-task skill distillation a unified RL problem, Living-Harness lets the harness itself iterate on failure experience, and TSDS equips edge agents with ’think short, defer smart’ calibration.

1. Latest arXiv Papers

1. SkillRise: Agentic Reinforcement Learning for Cross-Task Skill Evolution — https://arxiv.org/abs/2607.26784

2. Think Short, Defer Smart, Act, and Repeat: Calibrated Reasoning and Uncertainty-Aware Deferral for Edge LLM Agents — https://arxiv.org/abs/2607.26865

3. Embodied Agents Take Control: Minimal-Interface Zero-Shot Agents Rival Industrial-Scale Policies in Vision-and-Language Navigation — https://arxiv.org/abs/2607.26148

4. Living-Harness Is an Interactive-Agent Evolver — https://arxiv.org/abs/2607.26598

5. Mitigating Compounding Error via Video Representation Regularization — https://arxiv.org/abs/2607.27036

6. FreqForcing: Autoregressive Long Video Generation via Spectral Self-Anchoring — https://arxiv.org/abs/2607.27110

7. Swapping the Optimizer Boosts RL Agent Training Efficiency by Up to ~88% — https://arxiv.org/abs/2607.16169

8. Scaling GUI Agents with Visual State Transitions — https://arxiv.org/abs/2607.24112