ESR wrote a field guide for the vibe-coding era

What it does A living FAQ that sorts the signal from the noise in AI-assisted coding. Vilson Vieira and Eric S. Raymond map the current tool landscape—Cursor, Claude Code, Codex, Bolt, and open-source alternatives like Pi—and define the buzzwords (vibe coding, agentic coding, MCP) in plain English. It targets two audiences: developers who have not picked an AI assistant yet, and non-coders who want to build a SaaS by talking to an LLM.

The interesting bit The guide refuses to treat AI as a magic compiler. It explicitly warns that pure YOLO vibe coding collapses once project complexity rises, and pushes you toward “copiloting” where you remain the pilot as maintenance matters. That caution, paired with tactical advice like splitting work into PRD-driven tasks and running multiple models in parallel to validate output, gives it the credibility of people who have actually shipped software.

Key highlights

• Authored by Vilson Vieira and Eric S. Raymond, mixing hands-on practice with old-school engineering skepticism
• Distinguishes three modes: AI as copilot, AI as pilot (vibe coding), and fully agentic loops with orchestrators
• Recommends concrete tools for different skill levels, from web-based builders like Bolt and Lovable to local options like Pi and OpenHands
• Includes a practical prompting workflow: brainstorm, draft a PRD, break into tasks, and validate outputs across different models
• Warns that Claude Code’s API/SDK can burn through hundreds of dollars a day unnoticed

Caveats

• It is a curated markdown document, not code or an interactive tool, so its freshness depends on ongoing edits
• Some sections are essentially annotated link lists; the real value is in the framing and the workflow recipes

Verdict Read this if you are overwhelmed by the weekly flood of AI coding announcements and need a rubric for choosing tools. Skip it if you already run a 24/7 agent swarm and your billing alerts are silent.