The money sink is the problem, mostly because it appears to be yet another exponential increase, like the population approval thing. This limits the number of agents you can purchase for all practical purposes to a rather small number, depending on what you consider to be acceptable spending. True, you can spend 25% of your income on agents, but who can afford to do that? And what do you get for it compared to what you can get by spending the same money on other things?
This is the short version of my first experience with agents: Huge map, three oppnents, mid-late game. Every player has 20-30 planets with an average PQ of about 9. That gives >500 potential tiles which I can place an agent. By the time I got to agent 15-20, don't know the exaact number, I have to spend more than 10% of my income of about 4000 bc to get an agent every five turns. About half of the ones I bought went to nullify about one in four of the agents I got from the agent on every planet mega event. At this point, a new agent costs ~2000 bc and the per turn drain is almost equal to my wartime fleet of about 30 well armed medium sized ships. I noticed that the mega event chose no maintenance improvements first if there were any on the planet, so most of the agents were on banks. If freeing the bank gets me another 15 bc/turn, then it will take me about 130 turns or 2.5 game years to see a profit on that investment.
And that is just for the next agent. I have no idea what I would pay for agent #30 after another 10 increases in cost. All in all, I think the system is broken. There may be people here who can use the system as an effective weapon, but I strongly doubt that favorable circumstances to do this would present in every game, or even a majority of them. And even if you could win every game this way, who would want to? It would have to rank very high on the list of boring and passive ways to win. Frogby made a good case for not including multi-player in GC II, and I very much agreed with what he said. The devs for Neverwinter Nights II made the same arguments and I agreed with them too. They developed the game for the way most people actually played it, not for a vocal minority who live on Internet message boards. But isn't that what we have here? Maybe the beta testers like the new agents, a vocal minority if there ever was one, but from what I have seen that was by no means unanimous.
I think one one way to salvage the agents would be to allow us to remove agents by destroying the improvement they are on, if it can be destroyed. This would give us another option for dealing with agents at a cost that isn't game busting. I would think this would be easy enough to code compared to some other options. We would still be faced with nullifying agents on capitals and such, since they can't be destroyed, nor would you want to destroy them. Agents would be a costly setback, but not the bugbear they are now. Of course, rethinking the cost of agents would be nice too.
