Reading candidates 2026-07-14

[github-actions]

Reading candidates 2026-07-14

These links were collected automatically from curated RSS feeds.
Please review them before adding anything to reading/YYYY/MM.md.

• Window: last 7 days
• Max items: 24
• Max per source: 2

Candidates

1. Agentic Routing: The Harness-Native Data Flywheel

• Link: https://arxiv.org/abs/2607.11399v1
• Source: arXiv cs.CL
• Language: en
• Published: 2026-07-13
• Matched topics: llm, agent, coding-agent, infra, safety
• Score: 11
• Draft summary: Large language model agents are increasingly executed not by a single model call, but by an execution harness that manages observation, context, control, action, state, and verification. At the same time, frontier and open models are becoming structurally specialized: a model...

2. Agent Hacks Agent: Autoresearch for Production-Agent Red-Teaming

• Link: https://arxiv.org/abs/2607.11698v1
• Source: arXiv cs.AI
• Language: en
• Published: 2026-07-13
• Matched topics: llm, agent, coding-agent, safety
• Score: 10
• Draft summary: Production LLM agents such as Claude Code and Codex operate over untrusted content, files, commands, and workspace state, making safety failures directly actionable. Red-teaming must therefore keep pace with evolving models and tools. Existing approaches mainly optimize attack...

3. ToFu: A White-Box, Token-Efficient Agent Harness for Researchers

• Link: https://arxiv.org/abs/2607.11423v1
• Source: arXiv cs.CL
• Language: en
• Published: 2026-07-13
• Matched topics: llm, agent, coding-agent, infra
• Score: 8
• Draft summary: Agentic coding tools present new opportunities to transform research workflows. The performance of agent systems built depends on both large language models (LLMs) and the harness around LLMs, which is the orchestration code that determines an agent's behavior. We present ToFu...

4. NVIDIA Vera CPU Boosts AI Factory Throughput to Accelerate Agentic Workloads

• Link: https://developer.nvidia.com/blog/nvidia-vera-cpu-boosts-ai-factory-throughput-to-accelerate-agentic-workloads/
• Source: NVIDIA Generative AI Blog
• Language: en
• Published: 2026-07-09
• Matched topics: agent, coding-agent, rag, infra
• Score: 8
• Draft summary: Agentic systems turn model reasoning into action through multi-step workflows that combine inference, tool use, code execution, retrieval, orchestration, and...

5. Your coding agent bill doubled. Here’s how to fix it.

• Link: https://www.langchain.com/blog/fix-your-coding-agent-bill
• Source: LangChain Blog
• Language: en
• Published: 2026-07-08
• Matched topics: llm, agent, coding-agent, safety
• Score: 8
• Draft summary: Learn why coding agent bills spiral out of control — and how to trace, compare, and govern spend across Claude Code, Cursor, Copilot, and more in one place.

6. Using uvx in GitHub Actions in a cache-friendly way

• Link: https://simonwillison.net/2026/Jul/14/uvx-github-actions-cache/#atom-everything
• Source: Simon Willison
• Language: en
• Published: 2026-07-14
• Matched topics: agent, coding-agent, infra
• Score: 7
• Draft summary: TIL: Using uvx in GitHub Actions in a cache-friendly way I finally found a cache-friendly recipe for using uvx tool-name in GitHub Actions workflows that I like. The trick is setting a UV_EXCLUDE_NEWER: "2026-07-12" environment variable at the start of the workflow and then us...

7. When Local Monitors Miss Compositional Harm: Diagnosing Distributed Backdoors in Multi-Agent Systems

• Link: https://arxiv.org/abs/2607.11751v1
• Source: arXiv cs.LG
• Language: en
• Published: 2026-07-13
• Matched topics: llm, agent, eval, safety
• Score: 7
• Draft summary: As multi-agent, tool-using LLM systems are deployed, a common safety net is a runtime monitor that checks each message, tool call, or step on its own. We show this net has a fundamental hole. A distributed backdoor splits a harmful payload across agents, so every local check p...

8. Technical Report on the CVPR 2026@AdvML Workshop Challenge

• Link: https://arxiv.org/abs/2607.11560v1
• Source: arXiv cs.AI
• Language: en
• Published: 2026-07-13
• Matched topics: agent, infra, multimodal, safety
• Score: 7
• Draft summary: Vision-language agents (VLAs) are increasingly used to interpret complex driving scenes and support safety-critical reasoning. This report presents the CVPR 2026@AdvML Workshop Challenge on adversarial multimodal attacks against autonomous-driving VLAs. Built on DriveLM-style...

9. 🔥 SolonCode v2026.7.13 发布：多工作区协同、推理强度、任务分组

• Link: https://www.oschina.net/news/471686/soloncode-2026-7-13
• Source: OSChina AI
• Language: zh-CN
• Published: 2026-07-13
• Matched topics: llm, agent, coding-agent, infra
• Score: 7
• Draft summary: 1、关于 SolonCode（终端编码智能体） SolonCode 是由杭州无耳科技有限公司研发的企业级 终端编码智能体。它是一位全中文驱动的数字员工——能自主理解需求、自主规划步骤、自主编写代码。不挑模型，不挑平台，打开终端就能上岗。 核心差异化：SolonCode vs Claude Code 维度 SolonCode Claude Code 语言环境 全中文引导...

10. Reducing High-Bandwidth Memory Bottlenecks in JAX-Based LLM Training with Host Offloading

• Link: https://developer.nvidia.com/blog/reducing-high-bandwidth-memory-bottlenecks-in-jax-based-llm-training-with-host-offloading/
• Source: NVIDIA Generative AI Blog
• Language: en
• Published: 2026-07-10
• Matched topics: llm, agent, training
• Score: 7
• Draft summary: Large language model (LLM) training workloads increasingly run into GPU memory limits before compute is fully used. Model weights, gradients, optimizer states,...

11. Better tools made Copilot code review worse. Here’s how we actually improved it.

• Link: https://github.blog/ai-and-ml/github-copilot/better-tools-made-copilot-code-review-worse-heres-how-we-actually-improved-it/
• Source: GitHub AI and ML Blog
• Language: en
• Published: 2026-07-10
• Matched topics: agent, coding-agent, infra
• Score: 7
• Draft summary: How migrating Copilot code review to shared Unix-style code exploration tools reduced review cost by reshaping agent workflows around pull request evidence. The post Better tools made Copilot code review worse. Here’s how we actually improved it. appeared first on The GitHub B...

12. Claude Code 最强平替 —— GitHub Copilot CLI 上手指南，CC 深度用户

• Link: https://www.oschina.net/news/471618
• Source: OSChina AI
• Language: zh-CN
• Published: 2026-07-09
• Matched topics: llm, coding-agent
• Score: 7
• Draft summary: 为什么从 Claude Code 迁移到 Copilot CLI？ 最近 Claude Code 大量封禁，很多开发者突然失去了主力 AI 编程工具。有什么更好的替代方案？Claude + DeepSeek 组合做 Skills 自动化勉强可以，但编码能力差距明显 ——DeepSeek 经常出现编译错误、改问题不彻底，复杂重构基本扛不住。 经过反复尝试和对比，GitHub Copilot ...

13. Introducing Claude apps gateway for AWS

• Link: https://aws.amazon.com/blogs/machine-learning/introducing-claude-apps-gateway-for-aws/
• Source: AWS Machine Learning Blog
• Language: en
• Published: 2026-07-08
• Matched topics: llm, coding-agent, infra, safety
• Score: 7
• Draft summary: Today, we're announcing the Claude apps gateway for AWS, a self-hosted control plane that gives organizations a single point of control over access, cost, and policy for Claude Code and Claude Desktop. In this post, we show how to set up and run Claude apps gateway for AWS wit...

14. Launching UI for generative AI inference recommendations in Amazon SageMaker AI

• Link: https://aws.amazon.com/blogs/machine-learning/launching-ui-for-generative-ai-inference-recommendations-in-amazon-sagemaker-ai/
• Source: AWS Machine Learning Blog
• Language: en
• Published: 2026-07-13
• Matched topics: coding-agent, eval, infra
• Score: 6
• Draft summary: In this post, we introduce the UI for optimized generative AI inference recommendations in Amazon SageMaker AI Studio, a low-code no-code (LCNC) experience. The API already gives you programmatic access to recommendations, but it assumes you know which parameters to set and ho...

15. Auditing the Risk Claims of Distributional Reinforcement Learning

• Link: https://arxiv.org/abs/2607.11607v1
• Source: arXiv cs.LG
• Language: en
• Published: 2026-07-13
• Matched topics: agent, safety
• Score: 6
• Draft summary: Distributional reinforcement learning agents learn full return distributions that are increasingly read at face value: for interpretability, risk-sensitive control, and safety monitoring. We ask a question theory anticipates but that has not been measured directly: are the ris...

16. Directly Responsible Individuals (DRI)

• Link: https://simonwillison.net/2026/Jul/12/directly-responsible-individuals/#atom-everything
• Source: Simon Willison
• Language: en
• Published: 2026-07-12
• Matched topics: llm, agent, training
• Score: 6
• Draft summary: Directly Responsible Individuals (DRI) I went looking for a definition of "Directly Responsible Individuals" and the best I found was in the GitLab handbook. Apparently the term originated at Apple, where it's used to describe the person who is "ultimately accountable for the...

17. 连Claude Code都搞不定的巨型代码库，我们靠一个“自愈循环”给盘活了

• Link: https://www.infoq.cn/article/hSKvPpuMW3Y1GyyHtt3I?utm_source=rss&utm_medium=article
• Source: InfoQ 中国
• Language: zh-CN
• Published: 2026-07-14
• Matched topics: llm, coding-agent
• Score: 5
• Draft summary: 点击查看原文>

18. 从上下文到经验资产：Agent 记忆系统的工程化路径与 MemOS 实践

• Link: https://www.infoq.cn/article/RivZJwpYltVDJX5imMUX?utm_source=rss&utm_medium=article
• Source: InfoQ 中国
• Language: zh-CN
• Published: 2026-07-14
• Matched topics: agent, infra
• Score: 5
• Draft summary: 点击查看原文>

19. [AINews] Codex usage up >10x in 6 months to 7M users, +1M in the past ~day; did Codex overtake Claude Code??

• Link: https://www.latent.space/p/ainews-codex-usage-up-10x-in-6-months
• Source: Latent Space
• Language: en
• Published: 2026-07-14
• Matched topics: llm, coding-agent
• Score: 5
• Draft summary: a quiet day lets us fact check some numbers against the sound of silence of Claude Code reporting...

20. 正式开源！美团 LongCat-2.0 同步开放国产卡推理代码

• Link: https://tech.meituan.com/2026/07/12/LongCat-2.0-Open-source.html
• Source: 美团技术团队
• Language: zh-CN
• Published: 2026-07-12
• Matched topics: agent, rag, infra
• Score: 5
• Draft summary: 美团 LongCat-2.0 总参数 1.6T，平均激活约 48B，为真实的 Agentic Coding 任务而生，架构上创新性引入 LongCat 稀疏注意力和 N-gram Embedding，提升长上下文处理效率与 Token 级表示能力的同时，结合动态激活进一步强化了代码理解、生成以及执行的表现。

21. Improving Agents is a Data Mining Problem

• Link: https://www.langchain.com/blog/improving-agents-is-a-data-mining-problem
• Source: LangChain Blog
• Language: en
• Published: 2026-07-08
• Matched topics: agent, eval
• Score: 5
• Draft summary: How LangChain mines agent traces to find failures, fine-tune judge models cheaper than frontier LLMs, and hill-climb performance with evals.

22. Intelligence is Free, Now What? Data Systems for, of, and by Agents

• Link: http://bair.berkeley.edu/blog/2026/07/07/intelligence-is-free-now-what/
• Source: BAIR Blog
• Language: en
• Published: 2026-07-07
• Matched topics: llm, agent, infra
• Score: 5
• Draft summary: ... government of the people, by the people, for the people ... — Abraham Lincoln, Gettysburg Address (1863) The cost of AI is dropping rapidly. GPT-4-class capabilities cost roughly $30 per million tokens in early 2023; today the same runs under $1 , and some providers are pu...

23. 100+Skill导演级专家随叫随到！这回视频Agent终于有了可用级产品

• Link: https://www.qbitai.com/2026/07/449566.html
• Source: 量子位
• Language: zh-CN
• Published: 2026-07-14
• Matched topics: agent, multimodal
• Score: 4
• Draft summary: Agent携上百个Skills助我当大导演

24. Agent要数量也要脑子！浪潮信息一边单柜养4万Agent，一边让大模型组队答题

• Link: https://www.qbitai.com/2026/07/449311.html
• Source: 量子位
• Language: zh-CN
• Published: 2026-07-13
• Matched topics: llm, agent
• Score: 4
• Draft summary: 同时发布CPU原生液冷整机柜、多模融合超节点