Outsider to Agency in Creative AI Space

Background

I will briefly go into my background. Feel free to skip to “Long-term and Mid-term Goals” if you want to get to my plan for achieving agency in the creative AI space.

I came from an underprivileged background. My mom did not graduate HS, and education was not emphasized while I was growing up. My childhood bestfriend is currently in jail for unarmed robbery. The way that I initially pulled myself out of ignorance was through an interest in literature. This opened the floodgates to philosophy, music theory, art theory, math, etc. Well, maybe it was actually animation. I watched a lot of shounens when I was 12-13. I played a lot of naruto-arena (those who know know). As I outgrew shounens, I uncovered Serial Experiments Lain, Ghost in The Shell, Miyazaki, etc. I.e, works that were more than just entertainment. I then began to read Miyazaki. Perhaps due to his relentless allusions to classical music, jazz, and other authors, I very quickly found myself in a rabbit-hole. I will skip the details, but I very clearly had a deep desire to educate and culture myself, for whatever reason. This skips over some false-starts that I had, which I will also not get into.

I ended up going to Berkeley. Here I came across proof-based mathematics for the first time, and quickly became enamored. I took lower div math classes and did well (though it was an immense amount of work). Also, I found that my literature classes were not the best use of my time. If I wanted to become an author, I was better served dropping all of my classes and spending that time self-studying and writing on my own. At this point I was considering law school. I had a 4.0x GPA and a 99% LSAT, so I was close to a lock for a T14 law school, if not T5. For whatever reason, though, I felt the need to push myself to my limits. I took 20+ units of upper div and graduate level math courses for a couple of semester. In hindsight, this was stupid, but I guess that this was my way of figuring out whether I could achieve agency within math or not. I don’t really have an interest in working in a space where I don’t feel that I can be extremely exceptional, and this was my litmus test. The math professor who influenced me the most–Mariusz Wodzicki–would frequently go on rants, and many of his rants spoke about how to be exceptional in mathematics, you had to be precocious. If you were not a serious student of mathematics since the age of 12-13, it is unlikely that you’ll make significant contributions to the field. I for the most part agree with him.

So this lead me to move on from mathematics. Around this same time, a group of my friends had been accepted into accelerators and successfully raised a meaningful seed round. They had a similar background to I and my close friends, and thought that we would be capable of something similar. This was my unromantic entrance into the tech industry. Ironically enough, we interviewed at YC twice, as well as Techstars. We also received significant interest from other VC firms. Throughout this whole time, I was soul-searching. I didn’t really know what I wanted to do. Still, the tech industry, with its abnormal levels of meritocracy, deeply appealed to me. My impression is that it is an industry where agency and self-motivation are rewarded much more often than in other industries. I.e, if you’re good, you have a place.

And so I went down an immense amount of tech related rabbit holes trying to figure out what my identity could be within this industry, or whether I could even find an identity that made sense for myself within this industry. I was recruiting at this time. I got a job offer from a Series-B startup. I rejected it. The company did not really interest me, and at the time there was still a chance that my start-up would have been able to raise. I also found that I could not get myself to grind leetcode and toy implementation problems. It felt like a complete waste of time. In hindsight, maybe it is arrogant of myself to say this, but I intuitively felt that it was a low agency activity. Now, I very strongly feel that it is a low agency activity. It would likely lead to my either working at a big-tech company, or a LLM-wrapper startup, neither of which I have interest in.

Soul-Searching

So this was a year of immense soul-searching, where I tried to figure out what I should be doing, which was further complicated by the fact that I could not get myself to optimize for recruiting/interview-preparation. I built a couple projects. I thought that I’d be a build-in-public person. I thought that I’d start a youtube channel. My co-founder and I applied to a research grant (which is in its latter stages of consideration).

I ultimately concluded that what I had to do was find a way to leverage the 30,000+ hours that I’ve put into mathematics and literature/multi-disciplinary arts within the tech industry. This lead me to thinking about creative-ai. This lead me to think about what I would really want to do. What I came to realize is that I want to be a high agency individual within the creative AI space. I came to this conclusion in the first couple of weeks of June, 2025. It is currently July 7th, 2025. This was a major breakthrough.

Long-term and Mid-term Goals

And so, my life had been simplified to a single function: optimizing for agency within the creative-ai space. This gave me a significant amount of direction. I think that direction/self-understanding is the key to self-discipline, so this also helped a lot. I locked in (and would imagine that I still am locked in whenever this is read).

And so, how do I achieve maximum agency in this space? Yes, risk is a consideration, but I have always had an abnormal appetite for risk (being an author, hard-pivoting to math and dropping the prestige of law, etc). What I settled on is very concrete mid-term goals. I think that having extremely concrete goals is paramount when self-discipline is in question, so this is important.

My mid-term goal, which is ambitious, is to make meaningful contributions a combination of the following OSS projects:

• Exo
• Pytorch
• ComfyUI
• HF Diffusers
• vLLM
• Ray

My reasoning is twofold:

1. It is extremely important for my to create public artifacts of competence given my status as an outsider. I don’t think that it is moral to engage with the tech community if you are not a technical authority, or are not making an honest-to-god effort to become a technical authority.

2. I need deep intuition (from the hardware-software interface up to UI/UX) of the technology in order to envision the future creative AI products that I want to build.

I spent a lot of time trying to think of products that I could build (as part of my considertion of the build-in-public route), and my conclusion is that my knowledge of the technology is insufficient for me to conceptualize a product that I would be happy with.

A derivative result is that doing these things would also start the process of developing the network that I need in order to have agency within the space. I’d meet potential engineers, researchers, investors, and fundraisers that I could work with. This would probably be even easier if I were working at one of the companies that I would want to work at, but this route will enable me to network without being at one of those companies, albeit at a decelerated pace.

My justification for the OSS projects that I identified is straight forward. It moves from hardware-software interface upwards (exo/pytorch -> HF Diffusers -> ComfyUI). Also, low-level expertise will hopefully act as a insurance policy in the case that I cannot break into the field. There is a risk here: I have no interest in being a 99% percentile AI/LLM inference engineer (I’m fine with being 80%). What I aim to be is a 99% percentile fullstack/product engineer within the creative-ai/ml space. Also, for whatever reason, I am extremely drawn to the hardware-software interface space. I think that edge-computing and heterogenous computing are extremely interesting, and will be extremely relevant for the future, even within the creative/generative AI space, but that is a conversation for another day.

Short-term Goals

And so, having mid-term and long-term goals solidified, we move onto short-term goals. I think that this one is a bit more difficult, and will likely be subject to frequent iteration. Right now I am approaching things on a weekly-to-monthly basis. I am evaluating my rate of progress, and what I need in order to be able to produce the public artifacts that I need to produce. The one thing that I currently feel strongly about is working through Stanford’s cs 336 course, where I’m adjusting the curriculum to better align it with my interests (which will lead to future blog posts–small public artifacts). I’m quite certain that I can structure how I approach the course in a way that will lead to artifacts that touch a lot of the parts of the creative ML/AI techstack that I am interested in. That is sufficient for now. I’ll change courses if needed.

The above also touches on the other thing that I will do: I am going to start creating logs that track the topics that I am engaging in, and the projects that I am working on. I agree with Richard Feynman that teaching is an extremely powerful way to solidfy knowledge, and so I will take the time to explain the things that I learn, and also document my progress to achieving my mid-term and long-term goals. This will hopefully also lead to better self-awareness on my own productivity/progress.

As a parting thought, I think that with the advent of LLMs, learning has never been more democratic. As Yapcine says, “you can just do things.” Having someone who proves that you can in fact “just do things” would be incredibly empowering to the tech community. I hope to also add value by doing this.