How Our AI Agents Write a Help Center (www.veludocs.com)

A new AI-driven approach to automated help center creation has been announced, addressing the common issue of outdated documentation in software companies. Unlike traditional AI documentation tools that require access to source code, this system utilizes a login to navigate the product as a user would, gathering information directly from the interface. The result is a help center that includes written guides, screenshots, and videos, providing content that aligns closely with user needs, reflecting real-world application rather than code specifics. The process, which has evolved through several iterations, features a pipeline of AI agents split into different phases—research, planning, writing, editing, and capturing visual content. Each phase employs different AI models, such as Haiku for simple exploratory tasks and Sonnet for more complex writing and planning tasks. This structure ensures better accuracy and efficiency while controlling costs, as smaller models handle less critical tasks and the larger ones focus on judgment-based activities. By creating a consistent and visually coherent help center backed by well-organized mock data and targeted visuals, this method could set a new standard in AI-generated documentation.