> However I was shocked to learn that the game was not playable until very late
That seems rather common for many projects? If you have so many pieces developed in isolation, putting them together only really starts to happen when you get higher time pressure.
The part that's missing here is that Rome II is not a new project. The company has been working on this series of games for 25+ years now. They release them regularly and the bulk of the engine work was done years ago. The fact that they break the whole thing down and do it in isolation for every release is a mistake. They should just be iterating on top of the previous game.
> They should just be iterating on top of the previous game.
You can’t boot your whole game every time to do AI iteration. Whether or not the menu loads is completely irrelevant to your work. You don’t really want to integrate your broken shit with the currently working rest of the game either.
At least, that’s how I’d think about it.
You can always run the AI in a separate test harness to see changes more quickly. This is what they were doing throughout development. What they weren’t doing was putting everything back together to see if the game still worked or made any sense (until the very end).
If you don’t want to check in broken AI code then don’t; that’s what branches are for. But the fact that they weren’t maintaining a working master is inexcusable.
Even if it's common, I still don't think it's good.
Some games seem to be made playable way earlier then in the case of Rome 2. Look at Baldurs Gate 3 and soon for Path of Exile 2 - all the content is not there yet but the main "game loop" and features are there, this seems like a great way to get feedback, and have time to adjust if necessary. In the case of Rome 2 I guess the management imposed deadlines that did not allow this to begin with. The author does speak of a demo for E3 but it seems like a small piece that was artificially setup to look good.
This might be a partial explanation why indie studios can be in meaningfully competition with the AAA-developers. They often integrate early and use beta, even alpha, early access. As a layman I find that approach more understandable. In my work, I always favore incremental changes to large process overhauls. We do have to understand that 10 years ago early access was a lot less common.
In the specific case of Celeste, the game was playable before development, because it was preceded by the PICO-8 version.
Valve did this a couple times by just buying out teams who made mods or interesting games - Narbacular Drop becoming Portal, that goo thing becoming the goo thing in Portal 2, Counterstrike becoming Counterstrike, then a CS mod becoming Left 4 Dead.
"Every big program that works evolved from a small program that worked" is even true for games