How might Google Gemini's video analysis support software development?

The Scott Logic Blog offers insights, thought leadership, and technical expertise across various domains including software development, UX design, and financial services technology. Developers can explore articles on emerging technologies, industry trends, and software engineering best practices. Additionally, the blog covers case studies, project insights, and client success stories, providing  perspectives for technology professionals and enthusiasts.

Scott Logic

Google Gemini 3's video analysis capability was tested for exploratory testing in web development. A custom GitHub Copilot agent was built to automatically analyze screen recordings, identify bugs, generate reproduction steps, and triage issues against existing GitHub issues. The experiment showed promising results: the agent successfully identified visual bugs, mismatched data, and even unintended issues, while automatically generating detailed bug reports. However, it struggled with subtle issues like missing loading spinners and typos, and occasionally hallucinated bugs when none existed. The video analysis proved valuable for understanding state transitions and generating accurate reproduction steps, something screenshots cannot provide, suggesting potential for closing the feedback loop in agentic AI development.

Exploring Google Gemini's video analysis