After a few years of working with AI, I’ve come to see every form of AI assistance as a coin with two sides. Efficiency or laziness. Learning or dependence. Craft or access. Meaning or output. Each pair names a gap between what work produces and what work does to the person doing it. AI is the first tool that can hand you an output while taking away the process. Paradoxically, the process is often what has brought meaning and value to humans doing the work all along.

Note: In this essay I use the term AI to describe an LLM that has undergone some post-training process, not because I think it’s an intelligent system but purely because that’s been the colloquially accepted term for such a system (unfortunately)

Efficiency or laziness?

Suddenly we have a tool at our disposal that feels like having an assistant who will execute anything, anytime. AI promised efficiency, so we should be automating everything, right? Yes and no. Sometimes it’s faster to describe what you want and let the model build it. Sometimes it’s faster to do it yourself. Decide whether to be lazy or not, and don’t let laziness become your default. If you choose to use the AI to do the work for you, it still requires careful guidance. You have to deliberately design the work so that the assistant can execute it the way you intended. Otherwise, it’s going to deviate onto an unexpected path where you’ll spend more time cleaning up the slop than you would have spent doing the thing yourself. While sweeping the dirt off the floor, you feel useful and productive again because at least now you’re doing something. This can create a dangerous cycle of busy, unfulfilling work that’s anchored to an illusion of productivity.

In my early days of using AI to write code, I remember using it to run basic command-line tasks I would traditionally have run in two seconds myself. But the thought of typing in English what I wanted felt easier. In reality, I was being lazy and the network latency required to communicate with an LLM was higher than the human latency required for me to type the command myself.

Learning or dependence?

Every time you let AI handle something you could have learned from, you trade capability for convenience. Sometimes the trade is worth it, sometimes it isn’t. AI can help you work in domains where you have no expertise. It’s great at explaining anything you need to learn about along the way. But if you don’t understand the work, supervise it, and make the decisions, you’re building on a foundation you can’t inspect. It’s like never learning to multiply and always reaching for the calculator. As one psychiatrist who commented on an MIT study put it, “neural connections that help you in accessing information, the memory of facts, and the ability to be resilient: all that is going to weaken.” Once you learn, no one can take it from you. If you don’t, you’re blind without the tool. Create a habit centered on learning and discussion rather than auto-accepting machine outputs.

My first intro to AI as a coding tool was while developing a Flutter application. For the non-technical readers, Flutter gives you “legos” to build software. I didn’t know how to use Flutter and had limited web and mobile knowledge in general. During the AI hype, I kept seeing Twitter flooded with posts of people “vibe coding” a product and deploying it in 24 hours, so I figured I could just prompt it to do what I want and I’d quickly have a product up and running. I was wrong. It generated a lot of nonsense code and didn’t follow any basic software architecture. I was left with a mess to fix without knowing how to fix it because I didn’t take the time to learn what I was actually doing. Once I took a step back to learn the basics of Flutter, the process flowed much smoother. Don’t become a pilot who forgets how to fly without autopilot—Children of the Magenta.

Craft or access?

It’s expensive to hire human experts. AI offers access to skills without the financial cost. But at what creative and relational cost? AI is good at explicit, codified knowledge. It synthesizes documents, finds patterns, generates variations on forms that already exist, albeit rarely surprising or novel ones. Although it can come to a conclusion based on logical reasoning, we know that not all decisions are best made through logic. Some decisions are best made by relying on our gut feeling. Can we get to the point where we align human gut feeling with LLM outputs? I don’t know what that would look like but there’s a name for the thing that’s missing: Polanyi’s Paradox, “we know more than we can tell.” So much of human judgment lives in the tacit realm—in experience earned by struggling through the work ourselves or with others. If we cut others out of the work, we also lose out on a chance for human connection. A person can go for a walk, pass a plant that looks like the one in their grandmother’s backyard, call the friend they used to play with there, and arrive two minutes later at the answer to a problem they’d been stuck on all day. I don’t think AI is doing that kind of non-linear exploration, yet.

The next wave of technologists are going to need basic software engineering, product management, and design skills. I know I can do the first two well but I knew nothing about design, and after trying some AI tools, I got tired of the lack of texture and originality in the designs produced. I decided to read the book Graphic Design School by David Dabner to get a primer on design so I could create the website you’re viewing today. I talked to a couple of designer friends who gave me feedback and helped me along the way. Thank you. This site would have looked like every other AI-generated site otherwise.

Meaning or output?

The point of work was never only the output. It was always also what the work did to you: the identity it built, the competence it grew, the meaning it left behind. Beyond survival, work is one of the main places we answer “who am I” and “what am I good for.” When we optimize for output alone, we forget that work does something human beyond generating tokens and we forget that friction has value. The IKEA effect proves just this. People value what they build far more than the identical thing they didn’t, simply because they built it themselves. If you prompt an AI, take its output, and call it your work, will it ever give you what the work itself used to?

I make music as a way to escape the noise of the world and connect with myself. In making music, I’m after the process, not the output. I often have no idea what the output will be and only get a glimpse when I’m maybe three-quarters of the way through the process. For this reason, I don’t see the joy of creating AI-generated music. Sure, AI can help along the way. I’ve used it to help me create drum fills and give me chord ideas to help me connect two parts of a song in a theoretically sound manner. But asking it to produce the song for me is a waste of water and electricity.

So what do we do with this?

In this framing, I realize I treat the discussion as binary, heads or tails, where in reality there’s a continuous spectrum and balance that can be found in between the two options. Still, each moment of choice feels like a coin flip, even if the answer lives somewhere on that spectrum. I argue that using AI is neither good nor bad. It’s an act that asks for awareness. Maybe that’s the real skill of this era: deciding which struggles to keep. In a time where we can automate most white-collar work, the struggles we hold onto are the ones that still get to shape us.

Disclaimer: All ideas are my own. I used AI to help me in the editing process.