Why not just give your writers Claude or ChatGPT with the brand docs in a folder?
Try it first. For one writer and one client, it might be enough.
It stops working when the writing has to hold across a team. A project folder is a description of a voice; the model reads it and guesses. Five writers with the same folder get five results, and the model checks its own work, which is how tone drifts back in. What I build starts from the client's real writing, not a description of it. It runs the same way no matter who runs it, and a separate quality pass catches the mistakes your reviewers are tired of catching. And it's yours, not a feature of someone else's product.