Pre-launch update and reflections
My last update was rather positive in terms of how quickly things could get done, though it is now really approaching its final form, where assets are in place, many bugs have been squashed - but exactly that also remains the primary holdout, apart from the whole PR part of course. As you can see the UI has gone under several changes and based on feedback from the demo, it also features a proper tutorial. That was quite the oversight, as players often do need proper introduction to mechanics.
In total the game now utilizes no less than 7 custom effects and plugins, most of which already is or will be released for free on this platform. One of the real game-changers here, was an addon that can bake the tile-map into a texture, so as to reduce draw-calls and cpu hogging, while keeping the destructible terrain. In its current version, it can also update parts of the map texture, so not everything will have to be re-baked if only small parts are destroyed. Additionally and after much deliberation, the game also features a custom leaderboard, where players will be able to share their stats and compete for the top-spots.
The wishlist count on Steam is now 1.640+ which I consider a nice start, given there has not been done any outreach, but hopefully that number can be increased by some before the actual release (Which is not set in stone yet). One thing that still needs to be clarified though, is whether 
Tom would like to visit my private home again this year - or we good?
On the use of AI
During the course of the development I have used AI for a number of things, from delivering the base for small visual assets, translating the ingame and store text to other languages and even letting it give input on the whole codebase and writing parts (Like specific functions) for it.
For visual assets, I think its mostly recommendable to not wholely rely on it, as most models have certain styles and tend not to make things perfect or even downright copy proper art from real artists. At least some manual designing or editing is always heavily recommended for employing AI in visual art creation.
For translations, well, it does its job kind of decent, given enough context to the language files and it is a lot cheaper than paying for human translations, though again - use it with care and if possible, it should be proof-read by native speakers to avoid embarrasing translations, which can happen.
For code (JS, TypeScript, C3 editor snippets) I found especially Claude to be astonishingly competent in writing and delivering input. Things that would've taken me a day or two, could be done withing an mere hour of a little back-and-forth including testing said code. This is what I would most highly recommend to make part of your workflow, as it is also very capable of finding logic-errors and mistakes you could've made. Though again with at least a notice of care, as not everything AI does is correct.