The speaker outlines a personal workflow for building a set of AI-assisted tools, centered around the use of Claude for generating UI mockups and flowcharts through JSON files and Claude code. They emphasize Claude's ability to understand intent and be steerable, allowing for precise control over the design outputs. The process involves an iterative approach, starting with planning in Markdown, then using prompts to generate visual elements, which are refined with feedback before being translated into code. This method supports the creation of AI-native tools that can handle agentic loops and complex animations, ultimately increasing individual efficiency in design and development tasks.
In addition to design, the speaker uses Codex for code review, focusing on code alignment and refactoring suggestions to improve implementation quality. They express interest in AI-driven UI development, particularly in tools that allow for more creative and interactive environments, such as Google's Genie 3. The overall goal is to develop a flexible, personalized AI engineering workflow that integrates structured diagram creation with interactive design elements and code generation, streamlining the process of building complex applications.