I'm in a similar situation myself.
I have very little interest in mobile, and cross platform is only a minor bonus in my mind. My main focus is still Windows PC.
I've had agk for a few years, but haven't given it much serious attention until recently.
AGK has come along way in past year or so:
added plugin support, including a Steam plugin,
Built in JSON stringify/parse of arrays and UDTs (better arrays in general really, arrays in types, insert/remove/search/sort/length methods)
VR plugin (rift/vive :: I'd still like to see something targeted at daydream as well)
Janbo's shader pack is becoming comparible to AdvLighting (missing some things, adds some others), and opengl will likely have better forward looking support than dx9 with modern OS 's, though I'm much less familiar with glsl in general.
GaborD is also showing some really impressive PBR results.
Built in physics
Fast compiles
But it does some things in strange ways (from a dbpro standpoint)
You've only got 3 data types: string, float, integer (though it seems you can use int as a dword also)
Tweens, timing, 3d animation, window handling, screen resolution, and text are all a bit unintuitive or seem over complicated compared to similar in dbpro.
Many of the help docs have no code examples, but there are a good number of example projects available.
I haven't pushed its performance, it seems decent
but it is ultimately an interpreted language, so it's still a big question/concern hanging up in the air for me.
One of my concerns ous that it seems to have very few larger projects. Rush to Adventure comes to mind but that's about it.
Has anyone pushed it to the point where the compiler might start encountering limits and failures? Cumquat and Chris Tate have encountered line and function count limits in dbpro, I've recently encountered similar with a UDT that grew too large and/or too deeply nested (character module), and another where apparently too many UDT variables were declared in a single scope. (world module) I've had to refactor some things, and have begun to condense Boolean and byte UDT members into a fewer number of integers. This increases processing overhead and reduces readability, but simplifies the UDT structure presented to the compiler. I will likely end up moving constants and 'static' variables into my preprocessor to convert to 'hard coded' source before it goes to the compiler to further reduce the compiler load.
I just haven't seen any projects of a scale that would be likely to push agk to discover these types of limits yet. I have a decent feel for where dbpro 's limits are, I don't like the idea of moving to something whose limits are unknown.
I have been working on porting my framework over off and on, it's hard to get a good comparission until that is mostly up and I can run them more equivalently. The core and lower level utility modules have been converted, I'm getting ready to convert the UI module before long, but really I'm not in any rush, and my project takes priority.
I don't know, ultimately it hasn't quite proven itself to me as being fully capable of replacing dbpro for my main project yet, but it seems fine for maybe starting something smaller at some point.
It's easy enough to convert code from dbpro to agk in most cases, so i will likely keep with dbpro primarily, keep the framework synced over to agk at regular intervals, and keep my eye on how the language and performance progresses on agk's side as well as how problematic the compiler limits become on dbpro's side.
http://games.joshkirklin.com/sulium
A single player RPG featuring a branching, player driven storyline of meaningful choices and multiple endings alongside challenging active combat and intelligent AI.