What does bodybuilding have to do with AI? A lot actually. So much so that I named my new project kAI, after Kai Greene. TL;DR it's just a collection of tools that pipe various forms of video and transcriptions into a Google NotebookLM that I can talk to. It only cost me $16.64 in Google API calls to transcribe my 400 5-20min videos.

I have always been interested in lifting weights and figuring out the best, most efficient, effective plan to lift them. I'm never going to claim the title of “bodybuilder”, but I spent way more time on Bodybuilding.com and Tnation.com in my early teens and 20s than I'd like to admit.

I read way more about working out than actually working out [1]. Many orders of magnitude more. Mostly because reading is way less effortful than getting up, driving to the gym, sweating, coming home, taking a shower, and eating (refueling) appropriately.

One of the reoccurring, most unique characters on this pre-Instagram, early YouTube internet fitness world was Kai Greene. He was never a Mr. Olympia winner, but his nickname was “The People's Champ”.

From all the training footage bodybuilders of the era were uploading to early YouTube in the late 2000’s and early 2010’s, Kai Greene’s content showed us how much bodybuilding was an internal game to him rather than an external one. He let us in to the inner workings of his mind. Kai showed young, impressionable teenagers how profound this whole practice of lifting heavy-ass weights and eating tons of protein was to him, and could be for ourselves.

Of course, being teenagers, we just wanted the big muscles and drank the protein shakes that these bodybuilders were selling us. Reflecting back on this early influence, I can now see the wisdom that he was trying to get across as a fully grown adult. His muscles were an expression of his mind, hence, “thoughts become things”.

His ideas jive with me so much that I had been unknowingly making a project for years in anticipation of a technological shift that would enable it.

If you haven't already, you can join our Discord to connect with other like-minded AI-enabled software engineers and discuss the latest trends, tips, tricks, and techniques.

It's a great place to further explore topics that we discuss on the live stream and where I put all of the links that we pull up every stream.

Now back to the newsletter!

Around the time we moved to Las Vegas in 2018, I started taking videos of myself driving places and talking to the camera. I believe I livestreamed most of my drive from the Bay Area to Las Vegas on Facebook. Privately, in between trips to the gym and back, I talked about everything, all my hopes, dreams, and aspirations as a late twenty-something starting a new life with his then-girlfriend (now wife) in a new city.

In a lot of ways, I had already won. I was able to work remotely before it was cool. We moved to the fight capital of the world because it was just cheaper to live than in California. It also allowed us to enjoy all of the world-famous food, drinks, and shows. We were able to live where other people vacation. A dream come true in many ways.

And yet, I was still striving for more. I wasn't a senior engineer yet. I was an out of shape blue belt coming off of a knee injury. My then-girlfriend and I wanted to travel and see the world.

Fast-forward eight years, I have become a Black Belt in Jiu-Jitsu that has competed at a high level, a Senior DevOps Engineer with a focus on AI tooling, and married to the love of my life where we have taken not one, but two month-long trips to Europe in different years. We also own a home and have two cats. It's hard to see how far you've come without reflecting on things occasionally.

That's why I took all these driving videos, about 400 in total so far. I wanted to capture literal snapshots in time, over time. I figured at some point in the deep future, my kids could look up what their dad thought at a similar age to them. Somehow, that technology is here today.

If you have followed me for any amount of time, you have likely heard this analogy that we have been airdropped into the same small corner of a forest of low-hanging fruit with these AI tools. What most people do is they only look at apps and tools that they want to make from a purely software developer perspective.

This results in everyone making their own version of an agent orchestrator, a to-do list, a habit tracker, etc. I'm done trying to convince people to not do these things because I think it's just part of the journey. Code is cheap enough to create your own version of a thing and not really set you back. It's just time that you're not creating something net-new that you uniquely can create.

This may just be a new version of messing around with your dot files and your personal terminal configurations. We set things up to our exact specifications and customizations so we can work better. There’s just a lot more code that we can write to do this now.

What I built for myself in kAI is trying to convert my recorded thought time, ~400 video journals over 8 years, into something usable. I have not revisited these videos at all. they have just been sitting in my Apple iCloud Photos collecting digital dust.

This was a project that I always wanted to do, but I was waiting for the right technology to come along… eventually. Little did I know I would be doing it so soon and implementing it myself!

This project is part of my personal grove of low hanging fruit: intersecting interests that I have cultivated over time. This grove is my unique perspective that I can bring to the forest, where only I can see certain low-hanging fruit. Instead of crowding around oversubscribed ideas, I can see and work on things that have much fewer competitors.

I’m able to pick this low-hanging fruit and relatively few others are in this space are thinking about it in the exact same way as me. There exists other apps that, if you record in their app, do all this note-taking and summarizing capability, but according to Claude, there aren't any apps that import an entire backlog of recorded monologue videos. Probably because it's relatively expensive per video and there are probably people with thousands of these types of videos.

I did this market research mostly to see if there was anything that existed so I didn't have to create it. I don't see myself creating a stand-alone app to sell in this space, but now this is a relatively unique resource at my disposal.

Being able to extract information out of this backlog of videos with the latest Google Gemini models and turning those into Google Docs, which feed into the NotebookLM, breathes new life into these files that I will likely never watch again.

This still kind of janky process works well enough for synthesizing what I have thought about and it can trace throughlines for how I have developed over time. I would like to think this is a way for me to compound my thought time. Or as the title may imply, compounding my unique perspective, aka taste.

If you’re new here, my thesis since January is that taste and distribution are the most important factors now that the cost of code has gone to zero.

Taste is determining what should be built. Distribution is who knows about it.

I have been compounding my distribution already. If you've been noticing my various social media posts about my GitHub code contributions, those are my Claude skills in action. That Claude skill killed my entire app idea for Roxas. I am having my own personal SaaS-apocalypse.

I'm also using Claude Skills to post snippets of my newsletters to social media to promote it. Next, I'm going to use the new Claude design feature to create slides that summarize my newsletter and post to social media as well. This eliminates a $10/mo subscription by doing 80% of what Gamma AI does for me but potentially even better since it's all integrated into my existing “posting to social media via Buffer” skill workflow.

Immediately after I got these distribution skills set up, I wondered, “how can I compound taste?”

Spoiler alert, one project like kAI isn’t going to define my whole taste. In fact, it’s only really useful at looking backwards in a very grounded way. It is terrible at trying to synthesize new ideas from your current ones. It just sort of rehashes what you have said already, but sometimes you just need a reminder and some encouragement to go in a direction.

It's sort of like having ChatGPT chat memory for the past eight years instead of two like most people. It's an incredibly personalized system that allows me to pull deep from my thought archives and remind me of things that I may have thought about already but haven't in a long time.

An extremely personalized example that it gave me was when I prompted it:

and then it proceeds to give me a very good outline that matches the title with references to my exact thoughts about work, relationships, and life.

Another instance was I asked for through lines throughout my life:

As you can tell, these are all incredibly accurate and grounded in my real, actual thoughts.

kAI is not some idealized Black Mirror-esque thing where somebody can jump straight to a particular day and time and see what I was thinking…yet, but it is a helpful tool for having an interface into my past so I can essentially ask myself ask more questions and potentially use those to connect the dots towards the future.

An even bigger spoiler alert is that the act of writing this newsletter is the thing primarily compounding my taste. I either write or voice dictate every word of it. It forces me to think and crystallize my thoughts into something that others can read. Writing essays in school worked the same way for learning topics. You have to put one word after another, one sentence after another in a logical progression that ends in a point.

My biggest takeaway is that since I've been doing these, I have come up with my own framework for understanding this AI era, and you should do it for yourself too. Do it in your own unique way from your own angle. We don't need more regurgitation. We have AI to do that for us.

We need more net new creativity in this world. AI is just a blender of what has existed already. We always need more human thought time working on problems and explaining why they matter.

[1] Avid readers will recognize my typical ecological dynamics bent of “knowledge of” versus “knowledge about” here. We can make the same point here regarding reading about the best AI tools and debating endlessly online about the most optimal way to use them vs. actually using them imperfectly and developing real skill + mastery over them.