Daily Research Brief 2026-07-16

📊 Token usage: estimated from retrieval and writing scale.

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

Editor’s Note

Today’s clearest thread: ‘agents are descending from cloud capability races to the edge and toolchains’. The Cyberspace Administration granted filing approval to 7 on-device mobile LLMs in one batch while enforcing new rules on anthropomorphic interaction — effectively issuing a mass-production license for ‘AI phones / native agents’, and Nubia’s NaviX Ultra and ByteDance’s Doubao phone put ‘system-level agents’ on shelves the same day — edge agents officially move from PPT to storefront. On the open-source side, the focus is concentrated on…

1. Latest arXiv Papers

1. Deep Interaction: An Efficient Human-AI Interaction Method for Large Reasoning Models — https://arxiv.org/abs/2607.14049

2. Do Agent Optimizers Compound? A Continual-Learning Evaluation on Terminal-Bench 2.0 — https://arxiv.org/abs/2607.14004

3. A Self-Evolving Agent for Longitudinal Personal Health Management (HealthClaw) — https://arxiv.org/abs/2607.13940

4. Experience Memory Graph: One-Shot Error Correction for Agents (EMG) — https://arxiv.org/abs/2607.13884

5. AgentCompass: A Unified Evaluation Infrastructure for Agent Capabilities — https://arxiv.org/abs/2607.13705

6. VideoRAE: Taming Video Foundation Models for Generative Modeling via Representation Autoencoders — https://arxiv.org/abs/2607.14088

7. M4World: A Multi-view Multimodal Driving World Model for Interactive Object Manipulation and Minute-long Streaming — https://arxiv.org/abs/2607.14005

8. DeepStress: Stress-Testing Deep Search Agents — https://arxiv.org/abs/2607.13920