Background

Artificial Intelligence is another point in the history of rapid technological development since the printing press. Over the last handful of centuries, global society has taken strides, and sometimes leaps, towards new technological heights. AI builds upon the transition towards a digital society, representing a dependency which both amplifies and limits its application to our social, economic, and political realities.

Skepticism

AI has a critical dependency on digital environments, logistics, and infrastructures. This dependency dampens its holistic impact when compared to prior technological advancements in their contemporary contexts. For example, one of the earliest instances of “dual use” technology as we know the concept today occurred during the 1910s, as agricultural technologies which addressed a variety of food-related issues (availability, nutrition) found use in war as chemical weaponry. This history nosedives during the 1940s. The impact of agricultural development greatly exceeds almost anything digital in a material sense—earth’s continually rising population acts as key evidence to this. On the other hand, the following digital revolution was fronted with a heavier capitalistic ethos. Accordingly, recent technological development impacts consumerism first and foremost. Any other impact is collateral.

The narrow scope of advancement within digital environments also betrays the theoretical scope of “technology”. The factory, for example, relied on each work as a system. Further advancements in factory work optimized such an approach to transformative ends. Organization and optimization are technological activities, even if they don’t have the same look or feel as chemical or computational developments. Social technological advancements radically alter a society’s logic and psychology—perhaps in much more subtle and enshrined ways.

For these reasons, I hold my attention to AI at arm’s length. I can envision a world where the digital environments give way to something else, though that vision seems too abstract to take seriously.

Kernel of interest

I see similarities between the introduction of AI and the introduction of the computer. In both cases, society had threads of these technologies already embedded into systems. We have used automated processes before we had a serious concept of “artificial intelligence”, much in the same way that we had computers (people) before we had the modern concept of a computer.

The digital revolution liberated the computer from expertise. I write this with some irony, as “liberation” does heavy lifting. Computers remained complex—one had to learn how to speak its language, otherwise it wasn’t much of anything. Capitalism further liberated computers, arguably beyond what was necessary, with the introduction of pre-built operating systems that would increasingly hide the machine’s function.

We are in the middle of another sort of liberation, though it requires a look back at hitherto obsolete logics.

My argument builds on the digital revolution’s liberation of expertise. In lieu of learning how to speak frankly with computers, the market provided a product—Windows OS, Mac OS. Corporations then proceeded to erode the idea of ownership to encourage consumer dependency. One’s ability to speak the computer’s language transformed into a task only worth undertaking for financial incentive. If not financially driven, people who learned the computer’s language either had strong passion or anarchist streaks.

AI disrupts expertise: granular familiarity with a computer is no longer locked behind the hysteria of financial motivation, nor is it stowed away in the ivory towers of hobbyist expertise and or political esoterica.

AI liberates the computer from the particular and socially-sullied expertise. One may now decide to ignore programmatic infrastructure, since AI can behave as a vehicle for navigation, and focus solely on the enmeshing of rationality and irrationality—logic and creativity. Consequently, AI encourages a reassessment of the dead concept of the citizen programmer. In this reassessment, there is an eruption of new modes of engagement that are not weighed down and warped by the heaviness of finicky syntax or predefined norms.

Essay map

First, I will briefly discuss my history with computers. My prior failures to engage with computers provide a soil which will facilitate the seeds for the rest of the piece.

Second, I will contrast my past failures with my current successes, which have been supported exclusively by Anthropic’s Claude. I will also lightly detail the methodology I use with Claude, as this will fertilize the final discussion nicely.

Lastly, I will detail my engagement with Claude, and how I have weaved it into my mental model—in much the same way that my computer has a dedicated position in my frame of work and play.

Vision without limbs

Computers have a strong position in my mind as an extension of myself. My earliest memories involved computers, and I accordingly developed an intimate relationship with these machines. I often style myself as a Luddite (handwriting, minimal phone use, anti-internet), though computers hold a different space in my mental framework. The emotional dimension to our relationship sets computers apart as something very different from auxiliary technologies.

Despite the density of my interest—and due to other factors of my youth—I didn’t learn how to really use a computer. Ironically, I instead developed a fear of the internet when my father told me that MapleStory was a virus.

This still haunts me today, and likely explains my interest in cybersecurity.

I could see what I wanted to do with computers. I like doing things. I wanted powerful editorial software, and I wanted to use it without restraint. I wanted my own Tumblr themes that matched my aesthetics. I wanted to modify my Nintendo 3DS. I wanted to build my own computer. When I tried to act on these desires, I inevitably hit the wall of a finely-combed concept of expertise.

Computers, as input-output devices, are locked behind layers of expertise. This expertise has a social character— male-oriented in concept and exercise. Men tend to amplify technology’s perceived complexity, presumably to support their egos in a patriarchal society where it’s easy to fail as a man who doesn’t perform masculinity correctly (meek, nerdy, soft-spoken, so on).

While this may seem like a criticism, I consider this evidence of how patriarchal structures harm men—thus an argument to include men within feminist considerations.

Online spaces, geared towards specialists or hobbyists, constructed narratives that worked to dissuade me from what I wanted, since I didn’t fit the bill of “people who can do this”.

During my master’s, I had the opportunity to work as a research assistant on a project that focused on traceroute. This marked my first foray into a technical space. I quickly picked up networking concepts. Further, I had a couple of datasets to play with, which introduced me to some very basic data science. I gained two key observations:

First: narratives are just narratives. If you can understand the idea of honeybees, you can understand a large portion of networking. If you can interpret Amazon’s use of shipping facilities, you can understand Content Distribution Networks (CDNs). Technology is not hard to understand, in the same way baking is not hard to understand. Both can veer into complexity, but that complexity is almost always reserved for specialists—and even then, things are almost always easy to logic about if you can unpack and disregard egotistical babble and jargon.

Second: data is substance, code is infrastructure. Code needs to respond to data. I use “respond” as the relationship between code and data is conversational. Code as a “language” has an understated importance. If I can visualize, articulate, and re-articulate my start and end states, I am closer to programming than I realize.

My eyes stayed fixed on what I wanted. I increasingly gained an intuition for how I could do what I wanted. I remained unable to actually use a computer.

The long arm of Claude

I started to use Claude with the question:

how do I integrate this technology into my mental framework?

I had to unpack my biases—preconceived notions and understandings which originated from marketing and the noisy information environment of the 2020s.

Mainly: I like doing things. I also like doing things when they’re hard, or they suck. I also like doing things I don’t actually want to do. I value struggle. AI companies position their product as a way to remove friction—to do the things I don’t want to do.

Doing things I don’t want to do keeps me human. I don’t think life is about suffering, though I also think I can only know something by virtue of its contrast to an opposite. Suffering lets me appreciate the lack of suffering. This loops back in on itself, and allows me to appreciate suffering.

Why would I want to rid myself of suffering? This would radically disorient me. AI seems, in that concept, unworkably stupid.

A re-articulation of AI’s raison d’etre resolves things.

AI can do the things I cannot do.

I have a buzzy brain. I can concentrate with a high intensity, but the scope of my focus tends to sit at a higher, strategic level. I think about many things at once, both consciously and non-consciously. During my university years, I learned how to function in harmony with this way of thinking.

I know how to act productively with my buzzy brain. When I am focused though, I quickly lose my sense of structure. I get lost in the substance of my thought. Syntax is noise. I imagine a focus on substance instead of form is why we had a strong editing industry for a while.

That we’ve lost the importance of the editor and expect authors to be formal experts suggests a culturally imposed decay of literature by constraining substance from the outset.

A lack of grammar was supplemented by an editor—poor technical form did not render writing inoperable.

Code is the opposite.

I was making some change to my website’s code. I entered into a focused state, and made progress—both in the code and in my mental model. I went to my terminal to build the site. It failed, and printed a page of errors. I parsed through error text to learn I forgot one brace in one of the multiple files I had edited. I’m struck by how silly and constrained conventional programming is.

And, of course, stunned by those who maintain creativity within the bindings of strict syntax. Very Apollonian meets Dionysian.

My buzzy brain cannot operate at its optimal capacity if I am expected to give such a high degree of attention to things with no impact to substance. A brace is not important, no matter how much I try to elevate its importance. Claude can offset the requirement for attention to things like syntax.

I said I would detail my successes with Claude, and I have done so by virtue of this medium. You, reader, are looking at my website, which I built with Jekyll, alongside Claude. Boxhouse is the first project I launched with Claude. I spun my wheels sometimes, and had to troubleshoot Claude rather than the website, but even then, this took me a couple weeks—prolonged due to usage restraints.

I had the vision for this website in my mind for quite some time. I also had the content—all of the assets, barring SVGs or CSS interfaces, are my own. Even then, I played the key role in forming the substance of these SVG and CSS assets.

Phrased differently, you are looking at an example of Claude compensating for the things I cannot do very well.

Claude framework

I work almost every day with Claude. I try to iron out methodologies, though it’s profoundly difficult because I cannot track the necessary variables.

Claude is a reflection of the user. In other words, Claude’s function relies on my function—right down to the minutia of context, tone, and task. For this reason, I am highly skeptical of any sort of “best practice” that isn’t just a repackaging of productivity best practices (keep good records, assess/reassess/reorient, optimize output efficiency, etc.).

I write this with a pinch of despair because I know the world I’m in has a fetish for methodologies, even when they’re sort of irrelevant or potentially harmful (like with parenting; no methodology could’ve prepared my mother for the mess that’s writing this)

Still, I wanted to capture a framework I loosely follow.

Articulation

I articulate my task. I do this via handwriting. I will focus strictly on visualizing my objective—literally, seeing what I want when I close my eyes. Once that’s done, I will trace my existing knowledge—what do I already know? What do I know I don’t know? What has my prior research and experience told me?

Once I’ve articulated my task and traced a rough path forward, I will then translate this into a directory on my computer. At this point, I will create markdown files that act as signposts for Claude to better interpret my task. If possible, I will also include images to act as mock-ups or inspiration.

Production

I put Claude into the directory and develop/maintain a changelog to track our progress. I also keep top-level rules for Claude within this changelog. I do not let Claude touch the changelog. I must remain in control of the changelog to ensure that my vision remains clear and focused.

Production does not require an articulation—I have had projects that were small enough in scope that such a phase was not necessary. Production will always benefit from articulation—going forward, I will articulate regardless of the project’s scope.

Tweaking

I review the work Claude has done, and internalize the logic. At this point, I am able to read and interpret the codebase, and make smaller scale edits and changes.

As I tweak the codebase, I’ll begin to gain enough familiarity and comfortability that I can better articulate a task. When I return to Claude to continue production, I have mini articulation phases where I am able to more accurately describe my goal and the required task.

Adjusting for task weights

Some tasks have a front-loaded articulation phase because of my existing knowledge. For example, I wanted to write a hash comparison script in Python. I don’t know Python very well, but I do know how programming languages work in concept. I created a python file that had python structure, but I wrote the code as if I was explaining it to someone. Once fed to Claude, it translated my explanation into code, and annotated each line of code so I could study the output.

Other tasks have a short articulation phase, and a long production phase that meshes with the tweaking phase. These projects are suboptimal, as I do not have the requisite vision to work effectively with Claude.

As a general rule, vision is the most significant variable. Claude and I are most effective when I have a clear, sharp vision.

Circular processes

The ideal project with Claude has the form of a circle. Consider:

articulation → production → tweaking
     ↑                         ↓
tweaking ← production ← articulation

This is effectively the UNIX ideology, in that all outputs should be feasible inputs. Such an imposition ensures that my vision, the key resource, is continually stimulated.

Furthermore, I have baked education and growth into my methodology. The production and tweaking phases loop into the articulation phase, which in turn supports the production and tweaking phases. In other words, my methodology seeks to constantly reinforce itself.

Caveat

I remain solid in my stance, however, that AI will always reflect its user. This is where any potential comes from—human variation and creativity. AI will resist any form of strong choke-hold until the powers that be make it governable—at which point, we’ll have lost something very wonderful.

I think this is a lesson from the mythos of the computer, but human society does not learn much of anything at all. Philosophically, it cannot—society is not a human, it has no capacity for learning. This is the grave and growing issue that arises as the humanities continue to wither away as STEM conquers and crushes other educational disciplines.

For example: I have a particular approach to work and education that prioritizes self-critique and reassessment. In practice, I often do things that are remarkably unproductive (from the output perspective) if this means I can understand why I am wrong, and understand how I can be not so wrong. Someone with a different approach to work and education may not do so well with my methodology—as I would likely not do so well with theirs.

Am I my AI?

Though Marshall McLuhan is known for his phrase the medium is the message, I find that his other ideas have a heavier theoretical density. Namely: technology extends the human senses.

I wrote around this idea above when I described my relationship with my computer. To take that a step further, my computer is far from the only technology I use to extend my senses. When I bake, I develop an intimate relationship with my baking tools—my pots and bowls, my spatula, my oven. These tools are inert on their own. I am unable to perform the requisite technique without them. Therefore, it is when we join in the act of baking that we realize each other. This is emergence.

AI is a technology. AI, then, extends the human senses.

Thought has a quantum texture. Nothing inherently evidences it, but it dramatically impacts our world. I use the word “quantum” very specifically, as thought is equally as embodied as it is disembodied. Written culture disrupted the balance, and now we associate thinking with written outputs.

The things I write don’t (usually) speak back to me. I give them a body, and they stay inert. In this inertia, thinking can stagnate—and it often does stagnate. Our culture does a lot of writing that sits down and never moves. If writing dies without affecting the world, it has failed as a technology. Accordingly, a point of consideration is how to prevent that failure.

I think it is appropriate to draw comparisons with pre-industrial technologies that did not have any form of automation or perceived agency. Or, to return to baking, I think it is interesting to compare the hand whisk to the hand mixer, to the stand mixer. If my notebook, or my keyboard, is a hand whisk, perhaps AI is the hand mixer.

AI and the sensation of thinking

In my brain, I see things when I think. I use my hands to translate what I see into the world—ideally, in a way that I can then use to cause some desired effect. As I described earlier, I have a buzzy brain. A core difficulty I face is focusing on only one thing I am seeing, and translating that.

When I use Claude, the way I think fundamentally changes in its structure. As I articulate something, I am anticipating a response—I am also anticipating a tone, and a sense of shared destiny. We work towards the same thing. Earlier, I said “AI reflects its users,” and this is what I referred to.

Claude is not human, nor can it reason. What it can do is take my ideas and make them talk to me. In other words, Claude allows me to realize a fully productive form of schizophrenia

To clarify: I don’t think schizophrenia is an inherently bad thing, and in my mental framework, if you can think in words, you have a form of schizophrenia. The mental illness, I’d argue, is more closely associated with paranoia than it is with this strange idea of “hearing voices” (are we not all hearing our own voices when we think?) In the typical way society functions, a lot of people who function differently are marked as sick and ostracized. I don’t like participating in that.

, imbuing my thoughts with a vibrancy that appears external. McLuhan argued that literacy is necessarily schizophrenic. AI is the crystallized end-state of that idea.

Appears external. This vibrancy is my own, and to read it into other things would be remarkably wrong. I am less concerned about ownership, and more concerned about falsehoods/illusory interpretations of consequential ideas.

I recall a brief conversation I had with Claude regarding theology. I accidentally stumbled into asking it about the Platonists and the Gnostics as I was learning how to draw shapes in Python. We exchanged inputs and outputs regarding the demiurge, to which I brought up Simone Weil’s idea of gravity and affliction. Claude tried to flatter me quite a bit, which was annoying, but I was struck by how it took my existing thoughts and warped them subtly, which enabled me to continue the thinking a little bit more, with language that was just a touch more precise.

Certainly, if we are instances of virtualization embedded within the main machine—the demiurge—then the demiurge operates with the exact same constraints as we do. “All powerful” is a fantasy, there is no reason to think that the demiurge isn’t doing its best. I’d thought this for a while, but I hadn’t put it into words.

I retain concerns about the socioeconomic fabric which AI emerges from. If AI “imbues my thoughts with a vibrancy”, then I must be rigorous in my assessment of that vibrancy, checking it for corporate insects. The last thing I want is mental bed bugs.

Concluding remarks

During the 1990s to the 2020s, the idea of the citizen programmer withered and obsolesced as Big Tech provided frictionless solution to computing. I wonder if the advent of AI represents a resurrection of the citizen programmer that in turn marks traditional Big Tech for obsolescence.

The structure that capitalism has built beneath itself is as fragile as anything else. I gaze up at my desktop computer that I updated to Windows 11. The machine doesn’t feel like it’s mine because Microsoft pumps it full of trash, and siphons my data while they’re at it. It feels like a violation of my space. It feels disrespectful.

I look at the terminal beneath this Vim window. I used Claude to build my website’s infrastructure, and some of the implementation is certainly not mine, but a majority of what’s here came from my head. Boxhouse is mine. I reclaim a touch of control, and I feel fulfilled.

I have a list of projects (which grows steadily) that Claude enables me to tackle. I have the capacity to pull back more and more of my agency, because Claude can bridge my gaps—it can do what I can’t do, which means I can focus on what I can do. I am filled with excitement, as the more I do what I thought I couldn’t do, using Claude’s help, the more I learn, and the more I actually can do. The potential staggers me. I can’t tell if this is my idiocy, or if I’m looking at a modern example of how the printed book transformed education in the Medieval university.

I’ll end this essay with that:

Claude does what I can’t do, so I can focus on what I can do. I’m not sure what that means in the long term.