LATEST INVESTIGATIONS
A source audit of Cursor’s multi-agent research on self-driving codebases, analyzing recursive planning, lock contention failure modes, and 1,000 commits/hour throughput.
A source audit of FSoft-AI4Code’s CodeWiki (arXiv:2510.24428), examining hierarchical decomposition, recursive multi-agent docs, and the CodeWikiBench evaluation.
A source audit of Anthropic’s Claude Opus 4.6, Claude Code Agent Teams, and the 16-agent C compiler experiment, examining task locking, Git sync, and multi-agent coordination.
A source audit of Google MCP Toolbox for Databases, evaluating connection pooling, IAM authentication, schema-aware query planning, and OTel observability.
A source audit of Composio’s awesome-claude-skills catalog, analyzing how standardized SKILL.md primitives and MCP turn LLMs into reliable enterprise operators.
A source audit of Qwen3-Coder-Next, evaluating hybrid attention MoE architectures, environment-grounded RL training signals, and agentic SWE-bench performance.
A source audit of enterprise Claude Code adoption, evaluating CLI agent architectures, dual toolchains, permission guardrails, and software delivery velocity.
An architectural comparison of terminal-native coding harnesses—from Gemini CLI and Claude Code to OpenCode and Copilot CLI—exploring why Unix primitives beat IDE extensions.
A source audit of 2026 AI coding patterns, evaluating Ralph closed-loop execution, modular skill registries, Git worktree swarms, and memory harnesses.
A source audit analyzing HKUDS/nanobot, examining its 4,000-line single-loop architecture, minimal tool routing, local vLLM serving, and agent minimalism.
Search article titles, categories, and tags. Full text is searched when needed.
Type to search articles.