Software engineer built ScreenKite, a native macOS video editor, in six months using AI coding assistants, spending over $130,000 in token costs.

The developer started the ScreenKite project on October 1st with Claude Code and Codex (code-generation AI systems), building a native macOS application with screen recording, multiple stream inputs, and GPU acceleration—capabilities previously requiring reference documentation that did not exist.

The workflow evolved to use Codex as the primary code writer with Claude Code and Gemini CLI for code review, and later rewrote the entire Mac application into a Windows version in roughly 58 hours using GPT 5.5 Medium with less than 7B tokens consumed.

The project now contains over 500,000 lines of code and exports 4K video 3× faster than Screen Studio; however, at least 30% of development was fully hands-off (from requirements to acceptance), while the remaining 70% still required a human in the loop for QA and product management.