What's stopping them from just moving away?
There are some mathematical forces that pull at the units to stay and not all of them can leave at all. There's something like a minimum of 2 of each property control unit that cannot leave (according to their own AI determinations anyhow - they might be able to be called by the broker system which I'm not sure how to turn them off from being able to answer to.) The more units that are answering to a need, the less the need is perceived in competition with the needs of the city they are currently in as well.
The property control units which a city needs should not even look for work in other cities. The other thing is that all those units contantly looking for work actually creates an huge slowdown.
If I trusted the brokerage system, that would be possible. I do not have enough of a handle on it to be able to make that the only means of controlling where they should be, particularly controlling how many of them get called to that job. I agree it's a little slow to process what currently exists. Furthermore, the brokerage also doesn't have a way to ask for transports to enable units on other landmasses to be moved across to where the need exists. The brokerage was never developed out that far. It also had severe problems until very recently when some of it got debugged by a very clever programmer who took a look at some settling issues. It MAY work a little more like expected after that and maybe there's a way to enact better control through it than there has been up to now because of those bugs. But due to them, and other limitations in the brokerage system, I found I could not enact enough measured control through that alone.
The city should have control over at least an minimal amount of property control units. Those should not look for work and not be able to move away in their own.
As stated above, that is established already, at least as far as I can. It is possible that brokerage calls are still pulling some out that shouldn't be able to and a simple trick to make them unable to answer to a brokerage call would fix that. With Crime Control, there is also an extra unit AIs on units that help to control crime that every city should want at least one of, the Investigator AI. Some should also be trained for use in visibility needs and maintained in each city. That's not QUITE working as planned and there are some things to still tinker with there to get it working right. But the point is that there's a much better anti-crime blanket being employed in all cities. Again there, the brokerage MAY be causing some problems I'm not anticipating by grabbing units that weren't intended to be used for the purpose it grabs them for.
This stuff is so astronomically complex that I need months of focus on it to make any real progress there and I'm tired of fiddling with it all. My job has complexified and become much more demanding and I'm mentally exhausted at the end of most days and wanting more to play on the weekends at the moment. I need some time to let go then come back and fall in love with the mod all over again - I know that's the process that is natural since I've gone through it more than once.
In the meantime, the AI is not crippled with how it functions at the moment, at least not until much later in the game. And thus I've been happy to rest on it a bit. Sure there's 5000 things that need to be improved about the AI. I think we all know that. A release can't imply we've achieved perfection. Just better by far than the last version and as stable as can be expected.
With all that said, if you can see ways to improve it, great! But are you SURE it should be destabilized BEFORE the next release? Many improvements in this realm of AI decisionmaking were accomplished just after the last release, much of which was addressing exactly what you're saying is still problematic, and I'm thinking it's best to work on this stuff just after a release rather than just before one. I know it's still imperfect, but it's a hell of a lot better than it was.