I firmly believe that for a game to have good combat dynamics, it needs enemies with strong, interesting game plans. You should be able to describe what enemies want to accomplish & how they go about it as simple rules of thumb.
To develop this further & make it interactive the game plans shouldn't be binary or one sided. Enemies shouldn't go from neutral to their victory or loss condition near-instantly. This makes their game plan simple and uninteractive - you'll force the player to react at the finish line instead of letting the player contest the enemy at every step.
This isn't so simple, though, because certain types of player mechanics leave enemies no room to develop interesting game plans.
There are 2 main ways enemies can challenge the player - either by exploiting their weaknesses (getting behind the player where they can't reach), or shutting down/limiting their strengths (super armor letting them ignore your move's properties). Game plans are inherently tied to exploiting weaknesses, because without that, enemies will become passive threats that merely hope to capitalize on the player's mis-input or bad reaction speeds. They simply won't have the room to act "smart".
I think that by focusing on creating a set of weaknesses, and exploiting them when designing challenges, you end up designing more exciting games. It forces you to think more holistically about every element in your game, and gives you more tools to balance challenges without shutting down or nerfing things directly.
A player's punch is always beating enemies because it has superior range & speed? Instead of reducing its range, or giving some enemies super armor, or even giving them ranged attacks, you could ask "what can't the move do?" It can't hit behind the player, it can't hit above, it doesn't move the player towards enemies, perhaps it's linear. Once the set of weaknesses has been identified or created, enemies can start playing "smart" to counter it - staying out of range, moving sideways to exploit the linearity of the attack, trying to get behind where they're safe from it, trying to jump around the player at the attack's range.
It avoids a kind of one-sided design that makes combat feel stale and forceful. The only downside is that players tend to react negatively to strong sets of weaknesses in games. It's something they have to get used to (& look at the game more holistically) in order to truly see its value. As a result you tend to have to balance these weaknesses carefully so players don't get too annoyed.
Another thing to consider is : you don't necessarily need to be too stuck on the "ground level" of action games, focusing too much on states, micro-spacing, etc. A gameplan can be more "meta" in a sense - using the play area as a whole to derive the player's weaknesses or strengths, and building an engaging game around that. Screen/play area control has its own layer of strengths and weaknesses after all, as shmups show.
Here's my attempt at defining natural vs artificial rules/challenges :
A game is a logical, interconnected system where different rules are tied with one another and affect each other. Every game has rules that cohere well with other rules (either by naturally following previous rules, or feeding back into them in meaningful ways), and rules that exist in a tense contradictory relationship with other rules.
I call the most coherent, inter-dependent sets of rules "natural", they tend to make up a game's fundamental rules (X/Y/Z spacing, hitbox interactions, state duration/limitations). While heavily contradictory, isolated and "meta" rules & challenges are artificial, usually system mechanics-style stuff such as meters, super armor, supers.
Think of it like a color wheel - when making natural rules, the goal is to branch off from a point slowly in a way where the line between the new point and the last is blurry and hard to separate. Artificial rules are like adding a whole new color on the opposite side of the wheel.
