Haven't written one of these in a while, but the most recent patch has caused enough consternation that I thought I'd write a blog post that I had intended to write a couple years ago. In fact, I thought I had written in, but searching through the archive, I can't find it.
At any rate, one of the main problems of setting up defensive PVP is understanding what your agent and your toons will do. There are two sides to that understanding, agent and toons. I once had the idea of breaking down toon "logic," but it proved to be too large a task, back when there were about 45 heroes (aka a long time ago in a galaxy far away). Rather than write an encyclopedia of rules for heroes involving weeks of code diving (maybe we can setup a group task at some point in the future), I'll just lay out a basic understanding. Then, I can get to the meat of this post, which is what I understand about the agent order of process.
Toons[]
Toons have set rules. These rules examine the battle situation in very basic terms (largely opponent class and some personal health or team health, with turn order also a possibility), possibly throw a random element, and go to town. Some moves have higher priority, some characters have a truly random spread of moves. One thing that was clear the last time I looked at the AI, is that the AI doesn't consider the class of the attacker, only the target. Thus, the AI will play an alternate classed hero (like Tactician Hulk) largely like he's a bruiser. Not necessarily a fatal game play error, but something to consider. Some toons have much stricter structure to their AI. Some have much looser. Some have bugs, where they do silly things. The best way to understand toon AI is to watch it against a variety of teams. As the AI will read classes, it is not sufficient to watch it against one team.
Agents[]
There are somewhere north of ten million possible agent setups, factoring uniforms, EISO, and 4 slots. Probably north of a hundred million. I'm too lazy to do the factorial. At any rate, due to the large number of possible agent setups, the agent AI is probably a bit more complicated than a toon's ai. I say probably, because I haven't found it (and haven't really given it more than a cursory search). At any rate, we can assume it will consider agent class, target class, maybe something about agent health, and number of remaining team members. We can also assume that it somehow looks at the weapons it has and prioritizes. We know the agent has always had a major love affair with the Monstrosity and the Hoarfrost Mace, using both turn after turn despite better options available.
So, the idea is to attempt to control the agent's actions with gear choices. A common thought was to use multiple quick action gears to force the agent to get to the key action. Something along the line of maybe Signpost, Scroll of Angolob, Neurotrope and Cube. That setup would ensure that, no matter which goofy order the AI picks, it will eventually wind up, on turn one, doing the Cube's L1 move. That was all well and good, but now, you cannot run multiple quick action devices as quick actions anymore, due to Stalled. Fortunately, there are other ways to skin the jerk who would use a metaphor that promotes violences towards kitties.
Quick Action[]
While the change to quick actions has hurt them somewhat, and made them less automatic to include, You can still get away with carrying one. For everyone complaining that they can no longer go Golden Heavy Ion Beam-Weather Control Device-Something-Something, they are forgetting that you can still use one. It's not what it was, but it's better than not having Quick Actions at all.
Special Functionality[]
Items with Special Functionality cannot be used incorrectly by the AI, largely because they cannot be used by the AI. You would not want to play four of them, as it would make your agent do nothing but refresh, but playing one will give you a useful doodad, while restricting the AIs choices. You can consider an SF device to be a new passive for your agent. Paired with a quick action device, you've now basically given your agent a passive, plus a quick action, leaving a 50-50 chance of the agent doing the thing you'd like to do.
Cool Down and Starts Cooled Down[]
You may look at items that start cooled down as a problem. I have always looked at them as potential solutions, a way to regulate the AI's behavior, to fit a particular strategy. Consider this setup, from deep in the archives.. .
The ideal version of this (and the one the AI is prone to use) is Scroll - Attenuator (can't use anything else in the first round, due to cool ups on SS and PF), then SS-PF. It could use the Attenuator again, except that for whatever reason, the item with a cool down or cool up is higher priority to use than the item without. That's the way to use two QA devices to solid effect, by using overlapping cooldowns.
A Word About Snappy Service[]
Snappy Service is not now what it once was. But it still lets you fire off Buff type actions that are not QA as QAs. While not as automatic as it once was, maybe you can find a way through using cooldowns and coolups, with snappy service, to open the range of devices and effects you can use to program your agent.
Putting it All Together[]
Consider all three concepts and you can tune your agent's actions very specifically. You have a wiki with convenient lists of Special Functionality gear, Quick Action gear, Buff gear (for Snappy Service builds), and every other item in the game for your cool down and cool up research. Let's see what you come up with.