This paper presents a multi-agent AI architecture that achieves 92.1% reliability by organizing specialized AI agents into teams with opposing roles and hierarchical oversight, similar to corporate organizational …
GEPA introduces a novel prompt optimization approach that uses natural language reflection and Pareto-based evolutionary search to optimize compound AI systems, achieving superior performance compared to reinforcement …
A randomized controlled study examining how AI assistance affects skill formation in software development, finding that while AI can improve productivity, it significantly impairs conceptual understanding, code reading, …
WebMCP is a proposed web standard that enables AI agents to interact with websites through structured tools instead of screen-scraping, providing both imperative JavaScript APIs and declarative HTML form annotations for …
A comprehensive guide to MicroGPT, a minimal 200-line Python implementation that contains the complete algorithmic essence of training and running a GPT model, including dataset handling, tokenization, autograd engine, …
RICHES presents a novel approach that unifies retrieval and generation by interleaving document retrieval with sequence generation tasks, eliminating the need for separate retriever and generator components in RAG …
Apple researchers present Ferret-UI Lite, a compact 3B multimodal language model designed for on-device GUI automation across mobile, web, and desktop platforms. The model achieves competitive performance through curated …
This research demonstrates that agentic keyword search approaches can achieve over 90% of traditional RAG system performance without requiring vector databases. The study compares tool-augmented LLM agents using simple …
Simon Willison explores how AI coding agents have made writing code nearly free, while good code still requires significant effort and new engineering practices. The article discusses the shift from expensive code …
Andrej Karpathy discusses the emergence of ‘Claws’ as a new layer on top of LLM agents, providing orchestration, scheduling, and persistence capabilities while highlighting security concerns with current …
This research paper presents six principled design patterns for building AI agents with provable resistance to prompt injection attacks, demonstrating their practical applicability through ten case studies across diverse …
SWE-Lancer introduces a comprehensive benchmark of over 1,400 real freelance software engineering tasks from Upwork worth $1 million USD, evaluating frontier language models on both individual contributor coding tasks …
SpeCrawler is a comprehensive system that leverages large language models to automatically generate OpenAPI Specifications from diverse API documentation through a carefully crafted multi-stage pipeline. The system …
