Tag
This article comments on the low barrier to entry for autonomous-evolving code agent tools, drawing an analogy to stand-up comedy performances, pointing out that in the current code agent field, everyone can claim to be skilled at complex tasks.
AutoDesign introduces a meta-harness optimizer that recursively improves a code agent for long-horizon structured media generation, achieving state-of-the-art results on paper-to-poster synthesis and outperforming commercial systems like Claude Design on the new PosterBench benchmark.
A single developer built a code agent harness in Rust that starts up 245 times faster than Claude Code (14 ms vs 3.4 s) and uses up to 20x less RAM, achieving 10.4k stars on GitHub.
Nanbeige4.2-3B is a compact agentic model using Looped Transformer architecture, demonstrating strong performance on agentic and reasoning benchmarks at the 3B scale, outperforming larger models such as Qwen3.5-9B and Gemma4-12B.
小米MiMo团队开源了编程Agent MiMo Code，基于OpenCode，针对长程自动化任务，围绕计算、记忆、进化三个主题设计了多种机制，如并行采样选优、独立完成度验证、动态工作流和循环记忆系统。
DataFlow-Harness introduces a platform that guides LLM agents to construct editable DAG-based data pipelines through incremental mutations, achieving high pass rates and reduced costs compared to script-generation baselines.
The author created an open-source tool called awman that implements Claude's dynamic workflows concept, allowing multiple agents/models to collaborate on tasks with features like leader-designed workflows, shared context, and automated remediation.
This post comments on the current state of the code agent field, pointing out that unfunded relay stations survive by reverse engineering and cracking, while funded relay stations develop multiple AI services. It also promotes the WanJie Ark Intelligent Assistant platform, which offers text-to-image, text chat, and other features.
A guide on building a reusable Claude Code Agent loop that can be pointed at different tasks like bug fixing, speed optimization, or cost reduction by swapping check scripts.
NVIDIA has launched SpatialClaw, a code-based training-free agent framework for complex visual-spatial reasoning tasks, achieving an average of 59.9% on 20 benchmarks, 11.2 points higher than the previous best model.
MoCA-Agent is a market-of-claims code agent that improves financial and numerical reasoning by decomposing questions into atomic claims and using specialist agents to buy/sell those claims, achieving strong results on multiple benchmarks using a fixed Qwen 3.6-27B backbone.
AutoPDE is a code agent that explicitly represents solver strategies for partial differential equations, improving pass rate by 14.2% over baselines on the PDE Agent Bench.
Open-sourcing html-anything, a tool that converts arbitrary data into high-quality HTML, supporting 75 Skills and 9 export formats, compatible with popular code agents like Claude code, Codex, etc.
Google Flow is a creative studio where creators use natural language conversations to have AI build personalized tools for custom workflows. The article introduces its underlying Gemini engine and code agents, as well as examples of VJ tools, character tools, etc. created in workshops.