If you have a line on a built O&C, then yes, get one. Moreso than the Pamela's, that will allow for more in the way of generative work. Pamela's, ultimately, is more of a timing source-type of device, and while it can do some of that, the access to those functions thru the UI on it is a little clunky. A lot of the point of modular is to have the controls...ALL the controls...right there to grab, hence the ginormous control panel on these things. Modular synths are literally a gigantic control surface, so the better you can control the process you've patched, the more expressive and complex and such you can be with them. If you've worked with VCV, you get a bit of this idea, but VCV without a touchscreen (which I have) doesn't exactly get the point across fully.
Generative music, basically, is a subset of a compositional school of thought known as 'process music'. The idea is to create a structure that approximates a certain algorithm, and let that run, sometimes while making adjustments. I've actually worked with the earliest algorithmic electronic instrument created, the Sal-Mar Construction, which was a built-from-scratch device created by one of my composition teachers, Sal Martirano, back in the early-mid 1970s. In the case of that device, Sal used analog computing hardware to control the behavior and interactivity of the control surface and the synthesizer subsystems, and that analog computer had to be hardwire-patched. So what happens when patching a generative algorithm is that the synth patches in a modular synthesizer are behaving much like an analog computer, generating and reacting to various voltage curve functions. But the synth(s) of the Sal-Mar are actually pretty simplistic sound generators; the real 'meat' in the instrument is the control surface interaction, manipulating the algorithmic structure as time in a performance passes.
So, ultimately, doing generative music requires an effective layer of modulation control sets, in which the 'control' layer is being manipulated by a 'modulation' layer, which in turn gets manipulated by a 'sub-modulation' layer, and so on, with the idea being that the generative process is non-repeating and controls a fairly restricted set of sound generation as the listenable 'foreground'.
Perhaps the best example of this is Brian Eno's "Music for Airports"; while this work doesn't use synths that play as part of the work's audible layer, the audible elements are on tape loops. These loops are 'modulated' by the decisions Eno made in terms of loop length, cutting them so that they have a great deal of mathematical difference with respect to the loop duration, so that once started, the loops will never 'line up' again. Then the 'submodulation' comes from random factors involved with the playback machines themselves, since very few tape machines are 100% precisely the same, a factor Eno also took into account. And while this sounds like it has the potential to be very chaotic, it's actually anything but that, because Eno's choices of material which was recorded on the loops was highly restrictive. Any one layer will always correspond to any other layer's tonality, degree of activity, and so forth.
Now, with a modular synthesizer set up for generative work, one way to do this is to have several very simple sequencers which are running at speeds that also avoid 'lining up', with the clocks for each of these also having some degree of nonrepetitive behavior, and another layer that modulates that slightly and/or reacts to control signals coming in from the layer 'above' it. However, once having devised this time scheme, then you have to be careful that the sound sources being controlled by this system are simple enough that a sizable coincident layer's sounds 'mesh', either melodically, harmonically, or texturally. Otherwise, it tends to sound like meandering crap.
That's the tricky part, one which takes loads of experimentation and experience with generating and manipulating generative systems to get results which sound really, really spot-on.
Now, what I would suggest is to listen to a lot of generative music first-off. Take careful note of what seems to be going on, and then try and analyze how that might be done with a system of control signals. Think about how the time in these works passes and how the composer arranged the generative system to cause the events, across time, to occur. Not easy once you start diving into it, but over time you start getting the idea of what the generative process behind these works might be.
Once you start getting a feel for that, you're going to notice that modulation sources are the key...not so much sound sources. So that, actually, is where you want to start. Look at CV modulation sources, how you can make them interact, interfere, and so on. Then, how do you get the actual sound controls out of that continually-shifting process? Lots of ways: multipling out control voltage curves, using comparators to trigger events, altering timing signals feeding sequencers, etc. Be inventive. And study for some time how to extract control signals from the generative function layer...there's loads of ways that the can be done. THEN worry about the sounds. And you're likely to find that the simpler you can make the sounds, the more effective a composition generated in this way can be, because they won't get in the way of each other and will interweave nicely. To see if it's right, let the result play for a long time...as in, all day, possibly. Hell, some of La Monte Young's pieces ran for many YEARS, as in his 'Dream House' drone installations.
Also note: never consider electronic instruments as being separate devices. If there's some conceivable way to connect them, then what you actually have is somewhat 'modular' already. For example, consider setting up a couple of MATHS in a way that you get non-repetition. Off of this, use a few comparators (devices that send a trigger or gate pulse when a set voltage threshold gets crossed) to 'read' some of the process's voltage curves. Then send triggers from those to...oh, let's say, the Volca clocks, all separately. Now you'll have non-coincident timing signals syncing each Volca, and if they're pitched in such a way that it all sounds 'right'...well, there's another way to do it, and proof that an 'instrument' in electronic or electroacoustic media is only bound by the limitations of your own imagination, not by which things are in what cases.
So...four days in, well, maybe you'll want to study the idea more closely. And longer. And don't just do that on MG, but actually take some real time to STUDY the idea by listening, looking into various algorithmic processes, examining how related concepts like analog computing and how chaotic mathematical processes govern 'organic' processes. I've been tinkering with these concepts for about...ah, 40 years?...so there's a lot of possibilities to mess with here, and the only limit is really imagination and how broad you can make that imagination. Definitely not solely a matter of what equipment to use.