To see the panels for the units I mention, go to prismata.net/units and use your browser's text search functionality. Depending on the browser, you can do this by pressing Ctrl+F and then typing text.
Lunarch has a mission in combat training now that introduces this concept, but I'll say it myself anyway. In some sets there may be high-HP economic units such as Trinity Drone in addition to high-HP attackers such as Gauss Fabricator. In sets like these, it might be a good idea to simply convert all of your Drones to Trinity Drones and build only the sturdiest attackers like Iso Kronus and Gauss Cannon, especially if the set lacks good defenders. This is called 'breachproof'. The reason you would do it is that you won't have to build defense. Instead, your breachproof units act as defenders, and they're usually a lot more efficient than actual defense. For example, Gauss Cannon costs slightly more than 7 gold and has 5 HP, but Wall costs almost 7 and has only 3 HP. Also, when playing a normal strategy, you will eventually have to turn to desperation sources of defense when the late game comes, which are less efficient than Walls. If you go breachproof, your "defense" doesn't get less efficient as the game goes on.
Sometimes you will see very skilled players purposefully allow a small breach. This is called gambiting. The logic behind it is that if it's only a small breach (often one for 0) and your opponent has to make significant sacrifices to actually deal their full damage, it might not be worth it for them to do so (or at least not worth it by enough that you should pay the additional cost required to defend fully). For instance, if your opponent has a Wall and a Rhino defending and a Drake as part of their attack and you gambit for 0, they would have to A) lose their Wall to save the Rhino, which makes defense much harder for them that turn, and B) give up a Blastforge through the Drake, and either spend five gold to replace it or make it much more difficult for them to buy defense afterward. Allowing a breach for 0 in such a situation is probably fine. It's especially relevant against large piles of Frostbites. Against say five Frostbites, it can even be okay to gambit a breach for a Tarsier or more - if they lose all five Frostbites, it's like they're giving you 15 free defense (minus the absorb they denied) next turn, which will give you plenty of leisure in which to rebuy that Tarsier you lost.
The first big key to set reading is to pick out key units, or units that can be the core of your strategy. Everything else is a support unit. Amporilla is a good example of a key unit. Doomed Wall is a support unit. So if your game plan is "I'm gonna go for Doomed Walls", then you're going to lose. But if your game plan is "I'm gonna go for Amporilla", you might just win. This doesn't mean Doomed Wall is a bad unit; on the contrary, it is an absorber better than Wall and therefore often a must-buy. What I'm saying is that it can't be your strategy. Generally key units are permanent attackers. Examples of other key units inlude Cynestra, Gauss Fabricator, or Drake. Militia, although it is a permanent attacker, is a support unit because it's not efficient enough to be your main source of damage.
The second big key to set reading is to pick out synergies with the key units. In a game with Iso Kronus, Hellhound, and Cluster Bolt, you want to go for Iso Kronus because even though Hellhound is by far the best unit of the three in a vacuum, Iso Kronus becomes a lot better with Cluster Bolt. The reason is that you can alternate spending your green on Iso Kronuses and Cluster Bolts, so that you get the full power of synchronized exhaust attackers (see 'Syncing exhaust attackers' under Tactics) without the drawback of having to delay your purchases.
The third big key to set reading is to pick out counters to units that would otherwise be good. In a game with Cynestra, Energy Matrix and Apollo, you want to go for Apollo because although Cynestra is usually great with Energy Matrix around, Apollo is very good against Cynestra (or really any expensive unit with 3 or less HP) because it can just snipe them.
Prismata is a very reactive game - the best strategy usually depends heavily on what your opponent is doing. Often a strategy that is really the best in the set can fail you if you mindlessly play it without looking at what your opponent is doing, because they might be playing a strategy that beats yours but loses to anything else reasonable. So be mindful of what you and your opponent are committing to with each move you make. For example, if your decide that going for Tatsu Nullifier is bad because there's no red support and an Electrovore rush will beat it, still think twice before going for a third Engineer, because once you do that, you lock yourself out of going for an Electrovore rush, and it doesn't matter if your opponent's strategy is beatable if you lock yourself out of the only thing that beats it.
Threat refers to units that provide the option of attacking, but often won't. One example is Drake: you buy it for the same price as 2 Steelsplitters, it attacks for 2 every turn, but it also gives you the option to sacrifice a Blastforge for 2 more damage. Normally this is a slightly bad trade, but if your opponent doesn't defend against that extra 2 damage, then it will cause a breach, making it usually an excellent trade. Therefore your opponent has to defend against the threat of your Drake click even though you won't execute it, because if they don't you will. Any chill unit that pays some cost to use it (such as Frostbite which self-sacrifices on use) are also a form of threat.
The best counter to threat in the game is vigilant defense, which means units that block while performing some other useful function. Xeno Guardian is the best example. Building an army of Xeno Guardians is usually a mistake, but against threatened damage, you should absolutely make an army of them and allow the opponent to kill some of them by using their freeze or threat - spending an army of Frostbites just to kill one or two extra Xeno Guardians in place of other defenders is not worth the loss of your threat. Urban Sentry and Borehole Patroller can function in this capacity as well, and Ossified Drone is also great for redundant defense. Also things that make energy useful (Cauterizer, Electrovore, Galvani Drone) are vigilant, because the Engineers can both block and produce energy at the same time. Finally, units that are prompt blockers but don't normally get any value from that fact are a form of vigilance - for example, Rhino. Rhino is a bad purchase for prompt defense and a bad purchase for just an attacker that turns into a defender after two turns, but if it can defend against threat on its prompt turn, survive and then go on to attack and block 2 real damage later, it's pretty good. This is also true of Shiver Yeti, Corpus, Feral Warden, Photonic Fibroid and occasionally Protoplasm.
Since red defense is so inefficient to compensate for red attack being so efficient, pressure units (like Plasmafier, Cluster Bolt, Grenade Mech) tend to do better against red strategies than they do against other green-blue strategies. This is a very general and widely applicable principle. It's especially the case against double-Animus units. Double-Animus units are the epitome of greed. See Amporilla, the attacker that can literally never be outvalued by anything, for an example. So naturally pressure units are the textbook counter to double-Animus strategies.
Units that can deny absorb (Lancetooth, Mega Drone, Bloodrager) support low-economy strategies in otherwise high-econ sets. This is because the biggest reason not to go low-economy is absorb, so naturally units that counter absorb lower the optimal economy size.
When you have a unit like Doomed Mech or Omega Splitter that can either attack or increase your absorb by defending, usually it's better to defend, provided it saves you from having to buy prompt defense immediately. The reason is inflation (see my math article): value that arrives sooner is better than value that arrives later, so you'd rather get BV (defense) now than take BV from your opponent half a turn later.
Attackers with exhaust are usually most effective if synchronized to attack at the same time. There are two reasons for this. First, many sets have no preemptive defense, which means the opponent has to either buy all of their defense on the turn they need it or else be penalized in efficiency (by effectively having buildtime added to their defense). However, they have a limited number of Blastforges, and so they can usually only buy one or two Walls per turn before they have to switch to less efficient means of defense. You can take advantage of this with units like Iso Kronus and Scorchilla because if the brunt of your attack only arrives every other turn, they will likely have to use lots of inefficient defense such as Rhinos on those turns. The other reason only applies if all of your attackers are synchronized this way, and it is that if you deal 0 damage on your off turns, your opponent cannot get any absorb on those turns.
When you're buying your first attackers, it's often a good idea to deal just one point of damage on your first turn of attacking. This way you make your opponent buy a Wall, but the Wall only gets half of its absorb value on its first turn.
Normally, when you're holding back a Drone to block with it, Forcefield is a huge improvement over that play since it gives you 1 more defense for only 1 green (and no gold because that way you get to click the Drone). This is true even if you only need 1 defense. But against threat, since your Drone won't die if you hold it, buying a Forcefield turns out to be the same as holding a Drone except for one difference: the Forcefield loses the option of returning the Drone to work later. If you don't Forcefield and you still need the Drone for defense next turn, you can just Forcefield then and you've lost nothing by holding it first. If you don't need the Drone for defense next turn, you can click it again and get more gold before Forcefielding it later. In those cases, it's clearly better to hold a Drone back.
Alright, that's all the general tips I got for you. Hopefully this helps you to stop making some of the most common newb mistakes.