I built a system that builds the system. 4 hours from idea to production. Here's my full AI coding workflow for Rails.

RubyFlow's platform is a resource for Ruby developers, offering insights into Ruby programming language, Ruby on Rails framework, and Ruby-related tools. Through articles, tutorials, and community discussions, RubyFlow offers insights into building web applications with Ruby and Rails. Developers can learn about Ruby gems, Rails conventions, and best practices in Ruby development to write clean and maintainable code.

Ruby Flow

A Rails developer shares a structured AI coding workflow called a 'harness' built on top of Claude Code. The system coordinates AI agents through brainstorming, spec writing, implementation planning, per-task reviews, and CI checks — preventing agent drift and technical debt. Key elements include convention skills that encode project patterns, hooks that enforce them, intent-level implementation plans, and a multi-step automated review pipeline. The result: a medium-sized feature goes from idea to production in roughly 4 hours, with the developer spending only one hour on QA and code review. The workflow is open-sourced as a superpowers-rails plugin.

My Harness: How I stopped babysitting AI and went kitesurfing

How do I use Claude Code, aka my Rails harness

<p>The spec-writing step before implementation is doing a lot of heavy lifting here. Without it, Claude Code tends to make reasonable-looking decisions that turn out to be wrong about the overall architecture — because it’s optimizing locally without a global plan. The per-task review step also resonates: letting the agent run a full PR-worth of changes without checkpoints is where most “agent drift” problems come from. Convention skills as encoded guardrails are a neat approach — basically the agent skill layer enforcing team norms.</p>


<p>Basically begging to be fired… admitting he/she are barely needed and will be needed even less as AI’s abilities progress</p>
