Vibe Coding and Existential Dread

March 3, 2026

Anything that makes my brother and I argue always weighs heavy on my mind for a long time...

Background

I've used AI in my coding workflow for several years now (crucify me). But the extent to which I've indulged in this technology has been pretty limited compared to the waters I've just waded into. I've pretty much only used LLM chatbot-y tools as a better alternative to search engines. Search engines have been becoming increasingly more impractical as the quality of content and information online continues to degrade. Or perhaps more accurately, the change is a result of the quantity of poor-quality content and information online increasing. There's more trash in the search results one has to sift through to find something useful than ever before. Search engines have gotten worse, and a significantly more convenient alternative has emerged.

This isn't really a coincidence at all, however. The alternative can be leveraged as a means to produce faulty, worthless search results to try and tap into SEO for ad revenue and other such monetization schemes, while simultaneously being the alternative to the now degraded search. I find the relationship between the two concepts fascinating, but I digress.

Copilot, OpenAI, and probably Gemini and probably Grok and brand x and whatever else can all take highly refined/highly specific questions as input and produce new answers refined to address the content of the initial question. This is much more convenient than the word association game that you must play to get search engines to point you toward the information you seek. It's been especially helpful in my coding workflow, because oftentimes in coding you will come across highly specific situations/problems without having any proper immediate guidance on how to go about moving past them. Research and learning become necessary to advance, and the AI chatty bot tools I have mentioned can often speed up the process of this learning and research quite a bit. It's not a system without its flaws, but neither were the ancient googles and bings of the past. But perhaps this new workflow of mine will be relatively short-lived. This level of AI usage is somewhat minimal and hands-off when compared to the new-fangled world of "vibe coding."

The Vibing Begins

You could probably classify me as a skeptic. I'm pretty slow to adopt changes and new technologies. I see most passing fads as passing fads and nothing more. So naturally, I got to my first vibe coding experience a little late in the game. I've heard of the idea and related concepts being tossed around here and there, but I didn't really pay much mind to them until I caught a bit of a ThePrimeagen stream where he was using a rather clean and intuitive looking TUI tool to proomt sloppy code into existence. This was my, still rather fresh, introduction to OpenCode. Any program that runs in the terminal instantly gains points from me.

OpenCode is an AI agent harness. It enables the LLMs it interfaces with to execute commands on the device it's being run on. Now, instead of being limited to making suggestions, the AI has the power to actually perform tasks. This means it can create directories, read files, write files, execute code, all that fun stuff. Of course, you wouldn't want to give AI the power to do this on a machine that is storing sensitive data, so running this program from a virtual machine is recommended.

I had some leftover Anthropic credits to burn through from another project, so I decided that I'd give this vibe coding idea a try to see what all the hype is about. I set up my environment and made my first proomt. It went something like: "Let's start a new project. Let's make a pong clone using the game engine love2d. This shouldn't just be a rudimentary pong clone; it should be a feature-complete game with a menu, options to play against an AI or another player, scoring, win/loss, etc. It should have a complete set of game features fitting for a modern pong clone."

Whenever I am trying to learn a new-to-me game engine or framework, I always start out by writing a pong clone. It seems like a solid and simple enough starting point to wrap my head around the stack's quirks and interworkings. I've written a pong clone in the Lua-based Love2D before, so I had a reference point to use to help me better understand whatever the agent creates. Besides, making pong is a very well-documented task, and Love2D is a well-documented engine, so the agent should possess a lot of context to reference when trying to complete this task. I wrote my proomt and passed it on to the machine.

All 387.44 million miles of AM's wafer-thin circuit layers started firing up before my eyes. Claude Sonnet 4.6 scratched its head as it ran into a 403 http error message while trying to research the software stack. "No worries - I'm familiar with Love2D well enough to plan this. Let me ask a few clarifying questions before presenting the full plan," it proclaimed. The proomted then proomted me. Asking for things like visual styles, win conditions, game difficulty settings, and sound design direction. I fleshed out my desires through our back and forth, till the agent and I seemed to share a vision for what it was going to create. I switched from plan mode to build mode and gave the machine the okay.

What it created off the bat was already a better pong than I ever had created. To be fair, I only used simple pong games as a benchmarking tool to compare different game-making stacks; I never tried to make a "feature-complete" pong before. But still, the machine had produced something of notable quality that would have taken me a day or two in mere minutes. It had cost me tens of cents in Anthropic API calls. I continued to proomt, until I had something more fleshed out and genuinely nifty.

All-in-all, this ended up costing me about $7 to produce. That seems like a lot considering that I definitely wouldn't be willing to pay $7 for a pong game of this quality were it being sold at that price. I'd maybe pay 99 cents at the most, but I've never really spent much time in the market for pong clones. At that 99-cent price point, ceteris paribus, I'd only have to sell 8 copies to make a profit. You can purchase a copy of my pong clone for 99 cents at its github marketplace page.

Ethics

Emphasis on the word 'my', because I'm really kind of unsure whether or not I can claim ownership here. For the whole project, I only opened up Vim and contributed a couple of small changes twice or thrice. Changes that Sonnet found, didn't recognize, and made the executive decision to immediately revert. I had to proomt twist its arm to get it to agree to keep my changes long term.

I know that there's already been some litigation over this exact concept. What party involved in the vibe coding process vibe owns the vibed code. Does Anthropic own my vibes? Do I? Does anyone? Historically, I don't think ownership has been determined solely by creation. You've always been allowed to draft a contract with a creator to establish some kind of arrangement that makes you the owner of the creation. Outsourcing the creation of something you own has always been fair game as far as I can tell. So, for the time being, without real set-in-stone laws about the topic, I think ownership comes down to the TOS agreement you make with the SaaS platform hosting the models. In many such cases, our generous AI overlords grant us ownership over our vibes.

But still, I feel like I can't, in good faith, take credit for this creation. I'll happily claim some degree of ownership where I can, but my contributions to the creative process feel like they have been somewhat minimal. At least in comparison to the coding projects I have taken on in my past. I feel like all I did was delegate/direct the project, I don't feel so much like a builder anymore. This puts me in an odd place philosophically, because I personally consider coding to be a craft to hone and take pride in. I think that the work that I produce reflects my skills and dedication to this craft.

My flaw has always been that I take greater interest in the means than in the ends. And the vibe coding process seems to pervert the means to an extent to which it makes it harder to take pride in the ends. At the same time, this seems like an amazing new technology; it would be silly not to leverage it where appropriately applicable. If I am to continue to toy with it, perhaps a shift in perspective is required of me. I'd need to become more ends-focused than means. I suppose such a creative configuration is probably already the status quo for most creatives anyways, it makes the most sense. But for whatever reason, my philosophy as a creative didn't naturally develop in this way. I've enjoyed making for the sake of making. The fun is in the chase, not the catch.

The Vibing Continues

I performed a couple more vibe experiments. I was surprised to see the agent parse through sibling directories of where I said to start a project to look for example code to reference without being told to do so. The agent would pull things it felt it needed off the internet, like a font it downloaded from GitHub with curl. For another project, it even synthesized its own sound effects using Python. I built a couple more games in a couple different frameworks. I found that the autonomous tooling performs best when working with popular frameworks that have good and readily available documentation. Naturally, it starts to slip up when working with niche software.

All these experiments cost money via whatever API I use to access the models. And while not being ultra expensive, I'm preparing myself for a kind of large purchase right now, and I really don't need another pit for burning cash in at this time. Thus, the costs have become a slight bottleneck.

I love the concept of being able to run LLMs locally for this process. That would relieve me of the dread that comes from being roped back into being a dependent customer of big tech brother. I am without a doubt, however, that I neither possess the hardware nor the funds to obtain said hardware to run local LLMs reliably. The implications of hardware becoming inaccessible at the advent of this coding paradigm shift are worrying. Imagine a world where practically no software ever gets written without passing through OpenAI or Anthropic's servers first. I don't think it would be wise to give that kind of totalitarian power to profit-minded big tech firms. The whole digital world would be under their thumbs till the next big thing comes along and shakes the Earth. Local solutions for autonomous coding are the only way to move forward freely and with liberty. How accessible such solutions will be in the long run, I can't say. I'd hope that the next wave of research to come out of the AI field deals with the idea of optimization and efficiency, finding out how to do more with less hardware, instead of continuing to force the demand and prices of hardware up and up. That makes for bottleneck number 2. It's an odd-looking bottle.

As a result of these bottlenecks, I have limited the extent of my experimentation for now. And therefore, I was not able to really push these technologies to their limit to see where they really excel and where their shortcomings lie. My current mental model for defining these points is something like this, performing common tasks that have been done over and over and written about and discussed in depth will be quite easy for AI coding tools to produce variations on. While the more unique tasks found at the cutting edge of the programming field, or the less documented niche tasks that go undiscussed in commonplace, AI will stutter and struggle to produce anything of any real value. Perhaps the technology will plateau, and this will remain the status quo. Or perhaps it will advance beyond this point, and all software developers and engineers will go the way of the long extinct species that is jr developers within the next three months.

The Existential Crisis Begins

I enjoy coding. I've spent the past few years of my life trying to develop skills in this area. But now, if we assume for a moment that this technology has not plateaued just yet, does it even matter anymore whether or not I have obtained these skills which I cherish so much? I'm sure this technology has affected the job market for making software. Openings for jr developers are in a momentary decline. I believe I am at a point in my coding career at which, if the autonomous programming paradigm shift were not occurring, I should be entering the professional work force. But I haven't put the time I've put in solely for the sake of landing a fantastical high-paying job. I legitimately enjoy writing code. I find the process fun, I find it challenging, I find it heart-breakingly frustrating at times too, but I always come back to it. I consider this to be my craft. If the rest of the world moves in a way to outsource my craft, it leaves me behind.

It's really too soon to tell if my pessimism towards the future of my craft is justified. I will continue to hone these skills and bang away on my keyboard, simply because it's what I love to do. If that eventually turns into something that becomes lucrative in some form that I am content to pursue, I'll reap the rewards of that. But otherwise, I'm here for the love of the game.