I like "intelligent" micromanagement, that requires strategic decisionmaking.
I dislike tedious "working the system" micromanagement.
To me the first is making policy decisions.
And the second is implementing them, a.k.a. pushing buttons.
I don't like pushing buttons.
I believe it would be easy to implement higher level policy decisions in an automatic worker by simply opening up the AI desion tree the game currently uses to determine worker actions into a chart (perhaps, terrain type versus possible improvements) and then letting a player fill in what they wanted on the chart.
Thus if a player (like me) pretty much always builds cottages on a flood plains, all they'd have to do is open the chart, indicate that flood plains should get cottages, and then an automated worker would always build cottages on floodplains.
Setting up the chart might take some time, but if the chart (or tree) were started with the AI's default decisions in place, it would only be a matter of tweaking the system. And if these charts could be saved as mods/options one would only have to do it once, or load up a "Best Practices" outline from a forum (instead of manually implementing the same as one is now forced to do).
As to exploits, the way to deal with them is remove them.
It is my understanding that one of the major problems (loss of efficiency) with automating a worker is that beyond simply getting the worker to do what one wants it to do (where and when) there remains the problem of controlling he EXACT keyboard click sequence of how the worker performs the action. And this EXACT keyboard click sequence matters, not because gamers enjoy clicking buttons to tell a worker to build cottages, but because it is built into the system.
For example:
Asked to build a farm three tiles away on a road, an automated worker moves three tiles, ends its turn, and then starts to build a farm. (2 turns, 1 turn building target farm)
A worker being manually controlled is moved two tiles, chops/builds/farms, backspace is hit to end the chop/build/farm at the end of the turn, and the next turn the worker moves one more tile and begins to farm the target tile. (2 turns, 1 turn building target farm, one turn doing something else).
Net gain, 1 turn of "work" for manually controlling the worker.
To remove this exploit, all that needs be done is implement partial turn credit for chopping/building/farming (a 1/3 turn chop = 1/3 of a chop). And without the exploit, the strategic need for manually controlling worker diminishes greatly. Which is to say, if the results were the same between manually controlled workers and automated workers, I doubt anyone would manually control workers.
There are other ways to streamline the efficiency of automated workers, not significantly more complicated than I have outlined. It's not a matter of programing skill, it's a matter of desire and intent, and since I am under the impression "opening" the code for players and making the game more modable is part of the plan for Civ 5, I can't see why something of this nature can't or wouldn't be done, either before Civ 5 hits the market or after by gamers.
Further, since the computer has an AI function for every aspect of the game (troop movement, placement, and fighting; city production, tile selection; tech queue; etc.), automating each of these functions should be just as easy for the human player.
One would not have to avail themselves of these automations, but one could if one wanted to, letting the game play itself and only tweaking and correcting habitually bad moves or decisions made by the AI. Undoubtably in time, correcting these bad moves would become known as exploits, as well.
per turn thanks to the hammer rounding.
(e.g. early game you could get +25% for connection to cap, +50% for harbor and the 4 trade routes from GLH and reaching Currency), each would get truncated to 1
iscretiztion title that popped up in my head because of this thread made me lol. 
