So I've been mucking about in the governors.xml in order to see how we can make the AI build in a better way.
There's a few things worth noting in the default gov.xml. First thing worth mentioning is that, while the AI does have 'specialist' worlds listed (research, influence, wealth, manufacturing and 'growth' are all there), these worlds all have the following at the bottom:
<FillerImprovement>Manufacturing</FillerImprovement>
<FillerImprovement>Food</FillerImprovement>
<FillerImprovement>Research</FillerImprovement>
<FillerImprovement>Influence</FillerImprovement>
<FillerImprovement>Wealth</FillerImprovement>
<FillerImprovement>Approval</FillerImprovement>
This means that, even on the worlds it's supposedly using for science, it will build a few research buildings and then fill up the remainder of the planet with random buildings. This makes a bit of sense presently, while the AI can't actually specialize spending plans, but is generally kinda weak.
On top of that, the AI seems prone to both upgrading buildings before building new ones (the most horribly inefficient way of increasing bonuses), and also insists on doing terraforming improvements very early in it's build queue, too; I've often seen planets with just 1 factory and no other improvements trying to terraform land and upgrade the factory to top level before it attempts to build anything else.
This is astonishingly bad play. If the AI is unlucky enough to pick up improved factories or soil enhancement, it immediately leads to it spending the next 50 turns plodding along with almost no manufacturing capacity while it attempts to build a factory, convert a useless tile, and then upgrade the factory. By the late game, this can be devastating; having to get the first factory to T4 and use all the terraforming can leave the AI with extremely low production bonuses for literally a hundred turns, while picking the lowest return on investment.
Could it be changed so the AI upgrades and terraforms last?