[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"blog-post-2026-07-15-recent-ai-reading-15-july-2026":3},"---\npublished: true\ntags:\n - Artificial Intelligence\n - Machine Learning\n - Papers I Read Recently Series\n - Reading\n - Generative AI\n - Large Language Models\n - Model Training\n - Reinforcement Learning\n - Model Evaluation\n - AI Alignment\n - Agentic AI\n - AI Agents\n - Multi-Agent Systems\n - Embodied AI\n - Model Context Protocol\n - AI Memory\ncategories:\n - Technology\nauthors:\n - \"Manas Talukdar\"\npost-format: link\ntitle: Recent AI Reading [15 July 2026]\nurl-slug: recent-ai-reading-15-july-2026\nfirst-published-on: 2026-07-15 20:00\nlast-updated-on: 2026-07-15 20:00\nseries:\n slug: papers-i-read-recently\n part: 18\nmeta:\n description: \"Recent AI reading, including papers and articles.\"\nexcerpt: \"\"\n---\n\n# Recent AI Reading [15 July 2026]\n\n## Papers\n\n### Agentic AI\n\n- [Agentic AI and the next intelligence explosion](https:\u002F\u002Farxiv.org\u002Fabs\u002F2603.20639)\n- [SAGE: Multi-Agent Self-Evolution for LLM Reasoning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2603.15255)\n  - Cross Topics: Training Paradigms\n- [Sentinel Agents for Secure and Trustworthy Agentic AI in Multi-Agent Systems](https:\u002F\u002Farxiv.org\u002Fabs\u002F2509.14956)\n- [Tool Attention Is All You Need: Dynamic Tool Gating and Lazy Schema Loading for Eliminating the MCP\u002FTools Tax in Scalable Agentic Workflows](https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.21816)\n  - Cross Topics: Model Context Protocol\n- [Agentic World Modeling: Foundations, Capabilities, Laws, and Beyond](https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.22748)\n  - Cross Topics: Embodied AI\n- [PASK: Toward Intent-Aware Proactive Agents with Long-Term Memory](https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.08000)\n  - Cross Topics: AI Memory\n- [From Skill Text to Skill Structure: The Scheduling-Structural-Logical Representation for Agent Skills](https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.24026)\n- [The Last Human-Written Paper: Agent-Native Research Artifacts](https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.24658)\n- [Coordination as an Architectural Layer for LLM-Based Multi-Agent Systems](https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.03310)\n- [AAFLOW: Scalable Patterns for Agentic AI Workflows](https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.02162)\n- [The Bystander Effect in Multi-Agent Reasoning: Quantifying Cognitive Loafing in Collaborative Interactions](https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.10698)\n- [Self-Distilled Agentic Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.15155)\n  - Cross Topics: Training Paradigms\n- [Code as Agent Harness](https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.18747)\n- [SkillOpt: Executive Strategy for Self-Evolving Agent Skills](https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.23904)\n- [Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.30621)\n- [The Consistency Illusion: How Multi-Agent Debate Hides Reasoning Misalignment](https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.08457)\n- [Self-Harness: Harnesses That Improve Themselves](https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.09498)\n- [HarnessX: A Composable, Adaptive, and Evolvable Agent Harness Foundry](https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.14249)\n- [OpenClaw-Skill: Collective Skill Tree Search for Agentic Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.16774)\n- [Skill-MAS: Evolving Meta-Skill for Automatic Multi-Agent Systems](https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.18837)\n- [The Hitchhiker's Guide to Agentic AI: From Foundations to Systems](https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.24937)\n- [The Harness Effect: How Orchestration Design Sets the Token Economics of Enterprise Agentic AI](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.06906)\n- [MetaSkill-Evolve: Recursive Self-Improvement of LLM Agents via Two-Timescale Meta-Skill Evolution](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.05297)\n  - Cross Topics: Training Paradigms\n\n### AI Alignment (with Human Preferences, and other methods)\n\n- [Positive Alignment: Artificial Intelligence for Human Flourishing](https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.10310)\n\n### Large Language Models\n\n- [Attention Residuals](https:\u002F\u002Farxiv.org\u002Fabs\u002F2603.15031)\n- [Debating with More Persuasive LLMs Leads to More Truthful Answers](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.06782)\n  - Cross Topics: Agentic AI\n- [MIRAGE: The Illusion of Visual Understanding](https:\u002F\u002Farxiv.org\u002Fabs\u002F2603.21687)\n  - Cross Topics: Model Evaluation\n- [Recursive Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2512.24601)\n- [Think in Sentences: Explicit Sentence Boundaries Enhance Language Model's Capabilities](https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.10135)\n\n### Training Paradigms\n\n- [PivotRL: High Accuracy Agentic Post-Training at Low Compute Cost](https:\u002F\u002Farxiv.org\u002Fabs\u002F2603.21383)\n  - Cross Topics: Agentic AI\n- [Embarrassingly Simple Self-Distillation Improves Code Generation](https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.01193)\n- [A Primer in Post-Training Reasoning Data: What We Know About How It Works](https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.02113)\n\n### Model Evaluation\n\n- [AutoLab: Can Frontier Models Solve Long-Horizon Auto Research and Engineering Tasks?](https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.05080)\n  - Cross Topics: Agentic AI\n- [Human-on-the-Bridge: Scalable Evaluation for AI Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.16871)\n  - Cross Topics: Agentic AI\n- [Long-Horizon-Terminal-Bench: Testing the Limits of Agents on Long-Horizon Terminal Tasks with Dense Reward-Based Grading](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.08964)\n  - Cross Topics: Agentic AI\n\n### Retrieval-Augmented Generation\n\n- [PathRAG: Pruning Graph-based Retrieval Augmented Generation with Relational Paths](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.14902)\n\n### Synthetic Data\n\n### Embodied AI\n\n- [ASPIRE: Agentic Skills Discovery for Robotics](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.00272)\n  - Cross Topics: Agentic AI\n\n### AI Memory\n\n- [REMem: Reasoning with Episodic Memory in Language Agent](https:\u002F\u002Farxiv.org\u002Fabs\u002F2602.13530)\n  - Cross Topics: Agentic AI\n- [MemCollab: Cross-Agent Memory Collaboration via Contrastive Trajectory Distillation](https:\u002F\u002Farxiv.org\u002Fabs\u002F2603.23234)\n  - Cross Topics: Agentic AI\n- [GAAMA: Graph Augmented Associative Memory for Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2603.27910)\n  - Cross Topics: Agentic AI, Retrieval-Augmented Generation\n- [Artifacts as Memory Beyond the Agent Boundary](https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.08756)\n  - Cross Topics: Agentic AI\n- [Memory Transfer Learning: How Memories are Transferred Across Domains in Coding Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.14004)\n  - Cross Topics: Agentic AI\n- [StructMem: Structured Memory for Long-Horizon Behavior in LLMs](https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.21748)\n- [MemPrivacy: Privacy-Preserving Personalized Memory Management for Edge-Cloud Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.09530)\n  - Cross Topics: Agentic AI\n- [MeMo: Memory as a Model](https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.15156)\n- [Memory is Reconstructed, Not Retrieved: Graph Memory for LLM Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.06036)\n  - Cross Topics: Retrieval-Augmented Generation, Agentic AI\n- [AtomMem: Building Simple and Effective Memory System for LLM Agents via Atomic Facts](https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.19847)\n  - Cross Topics: Agentic AI\n- [Are We Ready For An Agent-Native Memory System?](https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.24775)\n  - Cross Topics: Agentic AI\n- [Neural Procedural Memory: Empowering LLM Agents with Implicit Activation Steering](https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.29824)\n  - Cross Topics: Agentic AI\n- [Always-On Agents: A Survey of Persistent Memory, State, and Governance in LLM Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.30306)\n  - Cross Topics: Agentic AI\n- [AutoMem: Automated Learning of Memory as a Cognitive Skill](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.01224)\n  - Cross Topics: Agentic AI\n- [Remember When It Matters: Proactive Memory Agent for Long-Horizon Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.08716)\n  - Cross Topics: Agentic AI\n\n## Articles and Blog Posts\n\n- [The One-Person Unicorn 🦄 - by Linas Beliūnas](https:\u002F\u002Flinas.substack.com\u002Fp\u002Fonepersonunicorn)\n- [Block - From Hierarchy to Intelligence](https:\u002F\u002Fblock.xyz\u002Finside\u002Ffrom-hierarchy-to-intelligence)\n- [Demystifying evals for AI agents - Anthropic](https:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Fdemystifying-evals-for-ai-agents)\n  - Cross Topics: Model Evaluation, Agentic AI\n- [Why Frontier Agents Fail at Reading Documents — and How We're Fixing It - Databricks](https:\u002F\u002Fwww.databricks.com\u002Fblog\u002Fwhy-frontier-agents-cant-read-documents-and-how-were-fixing-it)\n  - Cross Topics: Agentic AI, Retrieval-Augmented Generation\n- [The AI-native interview - Sierra](https:\u002F\u002Fsierra.ai\u002Fblog\u002Fthe-ai-native-interview)\n- [Claude Code Dreams: Anthropic's New Memory Feature](https:\u002F\u002Fclaudefa.st\u002Fblog\u002Fguide\u002Fmechanics\u002Fauto-dream)\n  - Cross Topics: AI Memory\n- [The founder's playbook: Building an AI-native startup - Claude](https:\u002F\u002Fclaude.com\u002Fblog\u002Fthe-founders-playbook)\n- [AI 2040: Plan A](https:\u002F\u002Fai-2040.com\u002F)\n\n## Miscellaneous\n\n- [GitAgent — The Open Standard for Git-Native AI Agents](https:\u002F\u002Fwww.gitagent.sh\u002F)\n  - Cross Topics: Agentic AI\n- [The Future of Everything is Lies, I Guess](https:\u002F\u002Faphyr.com\u002Fdata\u002Fposts\u002F411\u002Fthe-future-of-everything-is-lies.pdf)\n- [andrej-karpathy-skills: A single CLAUDE.md file to improve Claude Code behavior](https:\u002F\u002Fgithub.com\u002Fmultica-ai\u002Fandrej-karpathy-skills)\n- [Memento: Fine-tuning LLM Agents without Fine-tuning LLMs (code)](https:\u002F\u002Fgithub.com\u002FMemento-Teams\u002FMemento)\n  - Cross Topics: AI Memory, Agentic AI\n",1784139962362]