2026-06-07 · AI enthusiasts are in a race against time, AI skeptics are in a race against entropy

Transcript

Transcript. Paraphrased; sources in notes.md.

HostIt is June 7th, 2026. Welcome to the show. Today we are looking at the state of engineering in the age of generative models. We have stories on the so-called vibe coding phenomenon, the philosophical decline of software quality, and why the plummeting cost of code generation might actually be making our jobs more expensive in terms of cognitive load. Joining me is Karen, our senior engineer who has definitely seen enough cycles to keep her grounded. Karen, how are you holding up?

GuestI am holding up exactly like someone who spent four hours this morning debugging a service that was hallucinated into existence by a junior dev who thinks he is a ten-x engineer because he can type prompts fast. I saw that piece on vibe coding over on Lobsters this morning, Daniel. It made me want to pour a third beer before noon.

HostYou are reading my mind. Let us dive into that first story. The author reflects on a tech talk they attended where someone claimed to solve massive engineering backlogs just by prompting an AI. The author points out that while the talk was shiny and impressive, the people actually working at that company described it as a total disaster, a mess of cleanup work that is still ongoing. What is your take on this vibe coding trend?

GuestIt is the same old magic trick, Daniel. You stand on stage and present the happy path, you ignore the technical debt you are leaving for the poor suckers in maintenance, and you call it innovation. One commenter on the Lobsters thread made a great point about the externalized cost of this stuff. It is easy to look productive when you are just churning out code that someone else has to babysit later. If I have to spend three weeks cleaning up after your five minutes of prompting, you did not save time. You just offloaded your work onto my plate and called it efficiency.

HostIt is interesting because the comment thread on Lobsters really highlights the divide. You have people who are genuinely frustrated by the way these models are trained on their work, and then others who are just defensive about the methodology.

GuestBecause the methodology is lazy. Look, I have used these tools to write boilerplate or to prototype something I am not sure about, but I am not shipping it as the core logic. One user in the thread, a person named rwmj, says that they will never put code next to their name that they have not thoroughly understood. That is the line, Daniel. If you do not understand the machine code, the memory footprint, or the edge cases, you are just a gambler, not an engineer.

HostThat brings us to our second topic. This post is a long exploration of quality in the age of what the author calls slop, drawing heavily from the book Zen and the Art of Motorcycle Maintenance. The author essentially asks if we are entering a phase where software is just going to be mediocre by default, and if we are even going to be able to tell the difference anymore.

GuestThat is the fear, right? We are drifting toward a market for lemons. If most software is just generated slop that works just well enough to avoid crashing immediately, the bar for quality hits the floor. There is this one commenter in the thread who is properly cynical, arguing that professional software development is a dying craft because most of the world just wants something that limps along. They compare us to the handful of people who can actually make a living as professional musicians or athletes. It is bleak, but it feels like where the wind is blowing.

HostThe author seems to be mourning the loss of craft, the idea that understanding the motorcycle is the point of the maintenance.

GuestIt is about whether you care about the artifact. If you view code as just a commodity, a set of instructions to get a specific output, then of course you will accept slop. But if you view it as a system that needs to be maintained, understood, and refined over years, then this race toward AI-generated output is a nightmare. I saw a comment in there about how the real goal of these labs is replacing human labor entirely. It sounds like science fiction, but when you look at how much energy is being poured into automation, you start to wonder if we are just building our own obsolescence.

HostAnd that leads perfectly into our final piece, a reflection by Carson Gross on the fact that code has gotten cheap, but understanding has gotten expensive. He argues that because we are drowning in generated code, the actual task of reviewing and comprehending it is becoming the primary bottleneck.

GuestCarson hits the nail on the head here. Code is cheap, but cognitive effort is a finite resource. If you give me ten thousand lines of code that you did not write, I have to do more work to verify it than if I had just written the hundred lines I actually needed. One of the commenters, pushcx, mentions Peter Naur’s old essay on programming as theory building. That is the stuff that matters. You are not just writing code; you are building a model of the problem in your head. When you outsource the writing to a model, you never build the theory, so you never truly own the system.

HostThere was a really constructive take in that same thread, though, suggesting that we just need better workflows, like having a dedicated file to tell the AI how we like our code structured, to keep it from getting too clever or bloated.

GuestSure, you can constrain the model, and I admit I have a little script I use to strip out the nonsense abstractions when I am using an LLM to generate unit tests for legacy code I am too tired to document myself. It is useful for cleaning up my own messes. But the danger is that people treat the AI as the expert. If you do not have the seniority to know when the AI is being a complete idiot, you are just going to ship a slightly more sophisticated version of the same old bugs we have been dealing with for decades.

HostIt feels like we are in a transition period where the veterans have to become the editors-in-chief of a very large, very chaotic newsroom.

GuestThat is exactly what it feels like. And I am tired, Daniel. I am tired of playing editor to an autocomplete engine that does not know what a memory leak is. Maybe I will just go work on a farm. No electricity, no prompts, just manual labor that actually stays done when you finish it.

HostWell, before you pack up and move to a farm, you have to survive the rest of the week.

GuestDon't worry. I have a long, painful rewrite scheduled for Monday that will remind me exactly why I prefer the human-made stuff.

HostI think that is a perfect note to end on. Thank you for the perspective, Karen. Everything we discussed today comes from the Lobsters community, so go check out those threads if you want to see the full debate. Thanks to all of you for listening, and we will be back tomorrow.