alt Hacker NewsI've also been making a game engine based on MonoGame (and prime31/Nez), and one thing I recommend looking into is headless mode for both the game and the editor. I started my engine in the first place because I want to get back into gamedev, but there's no way I'm going back to the convoluted editor-centered workflow of Unity/Godot. And with proper MVVM, you can get the game headless too, so agents can play without the UI and run thousands of simulations of the gameplay.
Overall, MVVM should be getting a renaissance because of how well it determines the layers and data flow for AI. I have multiple projects using it, and even in early 2025, models had almost no issues understanding my code and implementing new features. Now I set up MVVM-inspired headless architectures including for Swift or React Native projects and it works like a charm.
The thing is that doing a separation for editors necessitates that you create a "client-server" protocol and things needs to flow through it, and that adds quite a bit of complexity over a "plain" editor.
And for authoring tools, especially in gamedev that can really blow up in complexity, because you often want to create custom editors (imagine things like enemy paths, layered enemy paths, skin customization options, and so on) and every customized one risks adding more complexity to the protocol (unless ridiculously overspecified from start).
This is why "dear ImGui" is so popular (look at the example screenshots for the library.. tons of editors), the magic is that you can tie rendering of the editor to the editables quite easily so there is very little overhead in makin an editor (a task that normally can gobble up inordinate amounts of time).