Engineer Automates 90% of Front-End Coding with AI, Reveals Model-vs-Model Code Review Strategy

CJ Hess, an engineer at Tenex, has unveiled a groundbreaking AI-driven development workflow that significantly reduces the need for manual front-end coding. Hess claims his system now allows AI to handle more than 90% of these tasks, a feat he discusses in detail on the 'How I AI' podcast hosted by Claire Vo.

A key component of Hess's workflow is Flowy, a custom tool he built to translate Claude's ASCII-based plans into interactive flowcharts and UI mockups. He argues that this visual representation dramatically reduces cognitive load compared to working with text-based plans, allowing for faster evaluation and approval of design concepts. The episode delves into how Flowy streamlines the development process, making it more efficient and intuitive.

Hess emphasizes the importance of custom "skills" in maximizing the effectiveness of AI tools. By teaching Claude specific skills, such as generating proper JSON for Flowy, he has created a continuously improving ecosystem. These skills are tailored to his specific needs, making Claude increasingly powerful over time. According to Hess, Claude excels at understanding intent, making it particularly valuable for complex coding tasks where nuance matters more than raw processing power.

One of the most intriguing aspects of Hess's approach is his use of a model-vs-model code review process. He pairs Claude, which he uses for code generation, with GPT-5.2 Codex for code review. While Claude excels at quickly building features, Codex is better at identifying code smells, inconsistencies, and potential refactoring opportunities. This dual-model approach, he argues, results in higher-quality code than either model could produce on its own.

Hess also shared his insights on leveraging Claude, emphasizing that while GPT-5.2 might be "smarter," Claude's "steerability" makes it superior for intent-heavy work. This highlights the importance of choosing the right AI model for the specific task at hand. Additionally, he uses an alias, "Kevin," for Claude with bypass permissions, highlighting the increasing viability of "living dangerously" with AI permissions when proper Git safeguards are in place.

The 'How I AI' episode featuring CJ Hess provides valuable insights into the practical application of AI in software development. His innovative use of custom tools, model-vs-model code reviews, and a focus on intent understanding demonstrates the potential of AI to transform the way developers work. The accompanying resources provide actionable workflows for implementing these techniques.