agent

This article discusses how with the evolution of AI Agent capabilities, Skill (skill/workflow encapsulation) may replace traditional software as a more important application form. In the future, Agents will dynamically combine Skills around tasks, while software will retreat to the background as infrastructure.

Own the agent learning loop and rent the model; private evals, traces, and memory compound over time.

The article points out that when AI agents run in ephemeral sandboxes, git clone has become a new bottleneck (often taking several seconds). Matt Rickard's Corigin rebuilds the git read path to reduce clone time to 184-646ms, which is faster than sandbox startup (690ms TTI).

OpenKnowledge is an open-source knowledge base tool that combines the note-taking experience of Obsidian with the philosophy of LLM Wiki. It comes with built-in MCP, Skills, Chat, and TUI features, supports Git sync, and can complement Obsidian for efficiently organizing personal knowledge.

User @jakevin7 states they have deleted Superpower, believing that as model and Agent capabilities improve, complex skills are no longer needed and may even impair model performance.

The Maka project is undergoing a comprehensive pre-release review and has started compiling design documents, aiming to help developers learn Agent development. It is a local-first Agent workspace with 640 stars and an active community.

The author shares that short prompts work better when building Agents, emphasizing the need to clarify results, constraints, and autonomy, which reduces token consumption and rework costs.

普林斯顿博士后Shilong Liu提出自进化代理的三层分类体系：工件迭代优化、Agent Harness自我改进和无黄金答案的模型学习，系统梳理了相关概念和前沿工作。

The user introduces their content candidate workspace, which uses agent-reach and opencli to collect hot topics, and the Agent performs deep analysis to automatically generate documents.

An article describing how to reduce AI agent response latency by 1.7× by switching to a model that requires less reasoning time rather than focusing on decoding speed.

Kent C. Dodds is offering early access to Kody Koala for developers working on agents with MCP support to ensure compatibility.

Georgios Konstantopoulos recommends Incur, a TypeScript library for building agentic CLIs, highlighting features like agent discovery, session savings with fewer tokens, and token-efficient output.

A developer built an iOS app called Maha OS that uses the Model Context Protocol (MCP) to allow AI agents like Claude to monitor live Apple Health biometrics and trigger a screen-lock breathing reset when stress is detected. The post seeks technical testers to break the MCP integration.

OpenManus is a free open-source alternative to Manus, built by MetaGPT researchers in 3 hours, enabling end-to-end task execution via terminal with web browsing, tool use, and LLM support.

Crawl4AI is an open-source web crawler tool that converts web content into clean Markdown format, designed for LLM's RAG, Agents, and data pipelines. Zero API Key, fast output of structured data.

A comparison showing that an untuned 27B parameter model outperforms a tuned 75B parameter model in agent tasks, highlighting potential inefficiencies in scaling and fine-tuning.

Two Hong Kong students achieved a 5x speedup by adding another loop outside the original automated research framework, without needing a better model or more compute. It is considered one of the most useful papers for ordinary Agent developers.

Jason Liu open-sourced a personal monorepo template that gives Codex permanent memory and skills, acting like a chief of staff without a vector database.

OpenAI announces ChatGPT Work, a new agent powered by Codex and GPT-5.6 that can take actions across apps and files to complete work projects.

OpenAI announces the GPT-5.6 family including Sol, Terra, and Luna, claiming state-of-the-art performance across coding, knowledge work, and science with higher efficiency and lower cost.