Occasional sharing
This article walks through how to assemble a personal LLM‑driven wiki using Claude Code, Obsidian, and the open‑source qmd search engine, detailing the directory layout, CLAUDE.md workflow spec, ingestion/query/health‑check cycles, and how the approach can be extended for enterprise knowledge bases.
In March 2026 the AI‑tool ecosystem buzzed with new features—Claude Code’s Auto Mode, OpenAI Codex’s Rust overhaul, Gemini CLI’s ADK migration, and rapid iterations from Kimi and Qwen—but the real turning point was a shift from pure capability competition to a race for trustworthy, transparent, and controllable agent infrastructure.
The article analyzes Anthropic's guidance on building effective AI agents, contrasts workflow and agent architectures, outlines criteria for choosing agents, presents six incremental design patterns, and shares practical principles such as simplicity, transparency, and robust tool interfaces.
This article explains how to construct a functional coding agent by combining a large language model, a bash tool, and a message history loop, showing step‑by‑step code, system prompts, error handling, and a complete execution example.
The article analyzes Karpathy’s Autoresearch project, explains how its 5‑minute experiment constraint enables hundreds of automated runs, and details the author’s three‑layer OpenClaw pipeline that orchestrates search and analysis agents, discusses design decisions, pitfalls, and practical takeaways for engineers.
