So, I'll kick off.
My current SVN game is in early classical. I dominate my continent, but another AI is also entrenched there, and right now I don't quite have the force to remove it. However, the AI stats show other civs, who I have not yet met, WAY ahead of me (twice my city count, twice my research rate, ...), which obviously is a good thing from an AI-is-working-ok so far perspective!
However, the stats for the larger AIs seem to contain the seeds of their own destruction, and I predict they will implode sometime in about the next era. The reason is REV instability caused indirectly by crime. The AI is building all the anti-crime buildings and basically managing it fairly well, but as it grows it runs out of ways it knows how to use to manage crime, and crime eventually goes rampant once city sizes get into the 20s and 30s. The crime buildings this generates (once the crime level gets past about 250 or so) then produce instability the AI cannot cope with, and implosion results.
There are several tweaks that need to be made to the AI, but the BIG one (which I think will prevent the implosions, or at least delay them considerably) is that currently the AI knows pretty well how to manage crime using anti-crime buildings, but it has a very weak grasp on how to manage it using anti-crime units.
Since anti-crime from buildings caps at fairly low levels (there just aren't that many buildings you can build to fight crime), fighting it with units is the only way that actually scales to larger sized cities. As a result, not having a good grasp on how to do this, means the AI fails to control crime as its cities get larger.
Hence a new work item, which I will treat as fairly urgent (before V27) - improve unit-based AI crime management. That includes:
- More forcefully keeping anti-crime units in cities (it does know to do this, but only enforces it weakly currently)
- Actively building units to fight crime (right now it just knows to build good defenders, and part of the 'goodness' calculation for the defender AI includes property assessment, so crime fighting). However, it doesn't understand the direct link between crime levels and the need to build anti-crime units explicitly (so if it thinks a city is adequately defended it won't build more defenders, so it won't get more crime fighters indirectly that way)
- Allowing more criteria to be exposed to the unit contracting system, so that crime fighters can be explicitly requested