Agentic Coding

This is a comprehensive guide to Claude Code, agentic AI systems, and advanced workflow techniques that directly aligns with the user's interests. The guide covers Agent Harness Engineering for designing agent systems, Context Engineering for managing prompt and memory architecture, Loop & Graph Engineering for bounded autonomous workflows, and multiple specialized workflows including TDD, Spec-First Development, and Plan-Driven Development. It provides decision frameworks for choosing between CLI commands, skills, agents, hooks, plugins, and MCP servers based on task requirements, plus practical guidance on context budget optimization, memory systems, and cross-session state management. The repository includes reusable examples, security hardening procedures, team adoption strategies, and production reliability checks specifically designed for solo founders and small SaaS teams using Claude Code and similar tools daily. Advanced sections cover agentic software factories, agent teams coordination, code review workflows, and cost optimization through AI Unit Economics, with machine-readable indexes and an MCP server for querying the guide from coding clients.

The website content from martinfowler.com/bliki/ contains one article directly relevant to the user's interests: "Vibe Coding" (21 May 2026). This article discusses vibe coding as a technique where developers prompt LLMs like Cursor Composer to build applications without reading the generated code, contrasting it with agentic programming where experienced developers review and maintain the code structure. The article explains that while vibe coding is useful for disposable prototypes and rapid development, it carries risks around maintainability, security, and correctness. The piece also covers the distinction between vibe coding (forgetting code exists) and agentic programming (caring about code structure), noting that agentic programming represents how many experienced programmers now work with LLMs, getting them to write all code while maintaining attention to internal structure and quality. Additionally, "Interrogatory LLM" (14 May 2026) provides relevant context on using LLMs to create structured prompts and context documents by having the LLM interview humans with one question at a time, which relates to efficient agentic coding workflows and prompt engineering techniques.