There are a few general things we can say about agency across all TTRPGs. Probably the biggest of those things is "who gets the most consequential say over the nature and trajectory of play, players or GM, and how does system facilitate that?"
Then there are specific things we can say inherent to the transparency and potency of premise of a game, its design goals, and how well those are put into effect by the actual experience of the game engine run well and with integrity. So, for instance, take a game that has as a core component of play being "fight for what you believe in." One potential design is:
* Player-handle and conflict-charged situation-intensive. They are myriad, they are impactful, and the incentive structures of system yield a very nice balance of "catch-22-itude" as you move through situations fighting for what you believe. Say, you get boons that work toward advancement and/or perpetuate success, but the blade cuts both ways as all the fighting and boon-spending and fallout/failure (both the fiction of failure and currency related to it) just get you in deeper and deeper until play fully resolves...your character utterly crystallized by your "hand" in this unforgiving tug-of-war at the end of play. We know, without a shadow of a doubt, "what you believe in" because you've been forced to fold, raise, and push all your chips in relentlessly through the span of play.
An alternative design is:
* Player-handle-lite, conflict being intermittent with substantial downtime and pervasive recovery, and incentive structures that either don't interact with "fight for what you believe" or interact awkwardly/obliquely. The way the fiction changes and the way the world-at-large views your character (from "everyman" to patrons, powers, and cosmology) is at the moment-to-moment and accrued discretion of another participant who both judges you and plays your antagonists.
Now the first one of these would be something like Paladin play under Burning Wheel Torchbearer or Dogs in the Vineyard. The second would be Paladin play under AD&D. I would say its a trivial observation, with endless supporting anecdotes stacked on top of each other through decades, that AD&D is extraordinarily low-agency Paladin play when it comes to (a) the general question I proposed at the top and (b) the specific question of a player & table discovering who a character authentically is as they "fight for what they believe in."
This is just one easy example of how sensitive agency is at both the system-neutral, macro question level and at the design goal and implementation level within system. This is as true for the Paladin play paradigm above as it is for things like Social Crawls (can you navigate a nexus of NPC interests and conflicting-goals to extract what you want and/or minimize fallout?) or Dungeon Crawls (can you navigate a nexus of obstacles and their relationships to extract what you want and/or minimize fallout?). And each of those two can be systematized in different ways to achieve less or more agency. Finally, the question of agency isn't one of "which one is more serial, granular, atomized?" Its a question of "to whatever degree the crawl is elided, atomized, or abstracted, how well does the whole of the design fit together so that when implemented in play, both the moment-to-moment experience of the play and the throughline of the play is intimately and unmistakably connected to player inputs and an outgrowth of player inputs."
You can have serial, granular, atomized crawls that are absolutely rubbish in terms of compelling decision-point navigation, player inputs, play outputs being an expression of agency. And you can have the inverse as well. Its about the intersection of brilliant, systematized implementation of design goals and deft execution at the table by the participants.