Knut Melvær responds to Lee Robinson's post about migrating cursor.com away from Sanity, highlighting both agreement and key omissions in Robinson's arguments against building a CMS. Robinson's migration involved 344 agent requests and about $260, transitioning to a system using markdown files, GitHub, and Vercel. While he raises valid points about the complexities of headless CMS platforms—like cumbersome preview workflows and authentication fragmentation—Melvær argues that Robinson ultimately ended up creating a system that incorporates many CMS features, such as asset management and user permissions, even while claiming to ditch the CMS model. Melvær outlines the challenges that arise when managing content at scale. The assumption that a markdown file corresponds directly to a single page quickly becomes problematic. For instance, updating pricing information scattered across multiple markdown files complicates maintenance. When legal language needs revision across numerous pages, the lack of a structured system leads to potential errors and increased workload. He emphasizes that a CMS provides necessary tools for content referencing and management that markdown files simply cannot replicate as the volume of content grows. Git, which Robinson’s system relies on for version control, is designed for code, not collaborative content creation. Content editing often involves semantic conflicts that require nuanced understanding, unlike the mechanical conflicts seen in code. Melvær points out that teams using Git for content management frequently resort to workarounds like lock files and PR review bottlenecks, which undermine the very efficiency they seek. The reliance on AI agents to grep codebases for content misses the mark when it comes to more complex queries that a structured CMS can easily handle. Ultimately, Melvær argues that while Robinson's migration simplifies certain aspects, it introduces new layers of complexity that a traditional CMS effectively addresses.