Vercel's update, while framed as a seamless improvement, underscores a growing tension in the AI infrastructure layer: platforms are becoming arbiters of model performance, not just neutral pipes. The ability to push updated weights without changing model identifiers gives infrastructure providers considerable influence, potentially smoothing over model inconsistencies but also abstracting away control from the end user. In our view, this represents another step toward the platformization of AI, where the gateway service becomes as critical as the model itself, and versioning becomes a managed service rather than a developer-controlled variable.