It is all a matter of what they program them to do. If they are programmed to build improvements, I am pessimistic about their usefulness. I have been following the GC games for many years and the AI can't build improvements that will satisfy a human player. This is especially true with adjacency bonuses. I would trust the AI to colonize for me before allowing them to build improvements.
In a game today I set the governor to wealth on a PQ9 planet. It immediately started covering half the hexes with factories.
For the AI to succeed building improvements they would have to scrap adjacency, and just set recommendations for each hex. I.E. this hex is suited for factories, research, farms, etc. Then you could set the AI to build improvements in a certain order, I.E. build only factories until I say change. Even then based on my experience with AI in all startegy games, they will find a way to screw up.
In the old SMAC game, the only improvements worth building over the long haul were Forests, so you could set the governor to build only Forests and that worked pretty good. Bottom line, the more choices they have, the more they screw up.