Vibe Engineering
How to get the most out of coding with AI.
“Vibe coding”, Collins Dictionary’s word of the year, describes using AI to generate code through natural language. While powerful, it raises quality concerns. The solution gaining traction, and the term popularized by Simon Willison, “Vibe Engineering”or “Spec driven development” is all about adding engineering rigor by planning features before coding them.
The core principle: think first, code second. How much you plan is on you. Here are some frameworks I’ve checked out to help you find one that suits your preferred workflow. Some are lightweight and some more comprehensive,
0. Plan Mode
Most AI coding tools (Claude Code, Cursor, etc.) now include planning features. You describe what to build before generating code. Combined with context files (Claude.md, agents.md, rules files), this simple step helps minimize drift and clarifies intent.
1. OpenSpec (Fission AI)
Minimal documentation for existing codebases. Each change gets a folder explaining what changed and why. No complex processes, just clear accountability. Perfect for maintenance work and incremental updates.
2. Spec Kit (GitHub)
Four-step loop: spec -> plan -> tasks -> code. Forces clear definition of goals and constraints upfront. Language-agnostic, works with any AI tool. Great for individuals and small teams prioritizing clarity over automation.
3. Agent OS (Builder Methods)
Three-layer context system: Standards -> Product -> Spec. Each layer constrains AI behavior, ensuring consistent output across teams and agents. Essential for multi-agent systems or organizations needing standardized AI outputs. I got a lot of value out of this project.
4. Taskmaster
Breaks PRDs into dependency-aware task trees. Workflow: PRD -> task generation -> dependency mapping -> ordered execution -> progress tracking. Keeps complex projects organized whether executed by humans or AI.
5. BMAD METHOD (BMAD Code Org)
Simulates an agile team with AI personas (Analyst, Architect, QA, Developer, Reviewer). Each role produces structured artifacts for the next. Supports everything from rapid prototypes to enterprise workflows. Works well with organizations with a structured process.
Summary
Vibe engineering / spec driven development organizes AI development and improves output quality. But it shifts developers toward product management: less coding, more directing. Not everyone wants that change, but it may be necessary for reliable AI-assisted development.
I’m working throught that change myself but do find this approach a net positive. Let me know if you’ve tried this approach and how it worked out or if I’ve missed your favorite tool.