Taste, Rust, AI: Things That Keep Coming Up Lately
So What Are We Supposed to Do About It
Taste
RUST
AI
These are topics that come up frequently these days, as job security for programmers has grown increasingly shaky.
During the COVID era, companies hired programmers at an enormous rate, but as growth slowed globally and organizations began using fewer people, programmers found themselves unable to manipulate the external world outside their control, so these are topics they frequently turn to for a sense of internal stability.
Macroeconomic interest rates, corporate layoffs, and the frozen unit prices of the outsourcing market are not variables that individual developers can control. Given the nature of the IT industry, most of it, aside from infrastructure, amounts to producing luxury goods. B2B software is coupled to economic downturns β workforce reductions create demand for automation and cut costs β while B2C is fundamentally closer to luxury goods. And the vast majority of programmers build luxury goods.
When confronted with an uncontrollable external threat, the most common defense mechanism humans reach for is to retreat into a small internal world they can perfectly control.
The three keywords dominating developer communities these days β "taste," "Rust," and "AI" β appear on the surface to be discussions of technological progress, but when you dig beneath that surface, they look more like psychological safety mechanisms programmers have constructed to quiet a growing sense of dread.
Taste does matter quite a bit in programming.
Taste is the last remaining means of personal differentiation in a leveled market. Academic credentials, certifications, and framework experience all level out quickly. Everyone comes out of the same bootcamp, listens to the same lectures, and builds the same portfolio. But "good taste" has not yet been leveled. So taste discourse masquerades as aesthetic discourse while actually functioning as differentiation discourse.
Saying "I like this kind of code" is an elegant variation of saying "I am different from those people." I do think taste itself matters. I prefer to use food as an analogy for programming, and in this case bread-baking works well. There are many kinds of bread, but even with the same ingredients and the same type of bread, different results can emerge. So what falls within a given category (functions as a program), is reasonably good (performs adequately), and yet diverges within that space β that divergence is what I'd call taste.
Bread-baking has layers. The lowest layer is whether it becomes bread at all. If the dough doesn't rise, it's not bread; if the oven temperature is wrong, it's not bread. At this layer there is no room for taste. Burnt dough is not a matter of taste β it's failure. In programming terms, this means code that won't compile, crashes at runtime, or fails to deliver the required functionality. Someone who says "my taste is code that doesn't run" doesn't have bad taste; they have no taste at all.
The layer above that is whether the bread does its job as bread. Is it edible? Is the crumb moist? Did the crust brown properly? In programming terms, this means performance is manageable, bugs are not fatal, and requirements are met. Up to this layer it is still not the domain of taste but the domain of standards. Falling short here doesn't mean "a different style" β it just means undercooked bread.

(Baking bread dough with cookie dough placed on top β melon bread)
So what is taste, then? The layer where results diverge even with the same ingredients and the same type of bread. Crust thickness, crumb structure, salt placement, depth of fermentation β these divergences are taste. And these divergences only have meaning among things that have already succeeded as bread. The argument goes that since using AI mostly handles the part of making something that functions as bread, it is these deeper layers that end up distinguishing projects from one another.
Taking the discussion one step further, genre selection itself falls under taste. A baker specializes by area β crackers, soft cookies, sandwich loaves β and even within each area there are further distinctions. When someone who has been making cookies tries their hand at bread, something like a melon pan emerges, and within any given worldview the shape a bread takes grows out of the individual's own values.
In the end, taste is not unconstrained freedom but something closer to which of the many possibilities that satisfy the constraints you happen to prefer.
Of course, taste does influence standards to some degree. There are problems like choosing between code that is simpler but slightly slower versus code that is more complex but slightly faster, as long as it stays within the performance budget, so standards and taste are hard to fully separate. But keep adding minor exceptions and it becomes hard to form any general rule, so I think of taste simply as a bias toward certain possibilities within constraints.
That said, taste as described here encompasses many layers and many problems, and is in practice a question with no definitive answer. What remains is simply the question of how effectively one can display that taste to an audience.
Among the people who talk about taste, though, I think a good many of them care less about what taste actually is and more urgently need the impression that they have taste. Programming restricts degrees of freedom, and the differences that emerge within those restricted degrees of freedom are usually what gets priced as taste, so a programmer on the job market always has to market the superiority of their taste.

Earlier I wrote that "when confronted with uncontrollable external threats, humans retreat into a small controllable internal world." Rust serves as the most elaborately constructed version of that retreat. Rust is a memory-safe language and, because the borrow checker makes it very safe to use, people say it is well-suited to AI and is fast on top of that β but they leave out the fact that it still lacks proper cross-platform GUI support, that WASM performance under the workloads real products demand is not all that impressive, and that marshalling costs are real. Sure, people building IT infrastructure might find Rust a reasonable substitute for C++, but it is a bit baffling to see the same argument made by people building application software.
Since the AI era began, everyone has gone looking for advantages in their primary language and claimed it is the language best suited to AI. Python users say "AI knows Python best"; TypeScript users say "types catch AI's mistakes"; Go users say "simplicity is the virtue of the AI age"; Rust users say "the borrow checker filters out AI hallucinations." Each argument has some merit. And every one of them is post-hoc rationalization of their existing preference.
Personally, I have some doubts about whether AI-generated Rust code, with its Clone() calls plastered everywhere and the tricks deployed just to appease the borrow checker, is really that fast β but people love the powerful image that Rust projects. I'm not so sure, though. Would you choose Rust solely on the grounds that it is memory-safe, without understanding its worldview? Is garbage collection really that sinful? People say memory safety without GC is superior, but if you're going to reach for unsafe, I genuinely wonder whether Rust is worth using as a C++ replacement.
Memory safety does have genuine necessity in lower-level infrastructure. But there are not that many developers who actually work in that domain, so I honestly don't understand what there is to brag about when people introduce Rust into their own products without a solid understanding of the language.
People say compilation is safe because AI writes the code, but you still have to read AI-generated code β maybe not every line, but you have to read it β so why force yourself to work in a language you don't understand? Not many developers actually work in the lower-level infrastructure domain where memory safety is a real necessity. Yet watching people outside that domain introduce Rust into their products based on the single phrase "memory safety," without properly understanding Rust's worldview, and then brag about it β well, it makes you wonder.
Finally, AI is a different matter from taste or Rust. Taste differentiates my choices, and Rust makes my code safer, but AI replaces my labor. So the desire directed at AI runs in the opposite direction from the other two. Taste and Rust are desires about "becoming a better developer than others," while AI is a desire about "surviving."
More precisely, it is a survival question: "I must use AI so I don't become obsolete." As a defense mechanism against an uncontrollable external environment, AI is far more direct than the other two. Taste and Rust are retreats into a small internal world, but AI announces that the retreat has become a survival strategy.
AI is automation aimed at replacing white-collar office workers. I actually dislike the terms "white-collar" and "blue-collar" β I think they are simply labels the capitalist class uses to divide workers against each other. To borrow Malcolm X's analogy of the field slave and the house slave, it is house slaves being sorted into opposing factions. In any case, the automation machine is now beginning to roll over workers, and within that reality workers are searching among themselves for ways to survive.
In some quarters, people gather collectively to oppose AI, wanting to protect their rights and their current paycheck; in others, people rush to embrace AI wholesale, hoping to boost the productivity metric, outshine the rest, and keep their jobs. Both positions have merit. Either way, both are simply pursuing the preservation of their own labor value by different means.
Will companies actually pay out the gains in productivity as wages? And if you refuse to use AI, will the existing value of your labor remain intact? Neither future is clearly foreseeable. Everyone is just shouting because they're anxious.
Because it's the AI era, humans must do more creative work. Because it's the AI era, programmers must become architects. Because it's the AI era, taste matters more than coding. Because it's the AI era, safe languages like Rust matter more.
Each of these claims contains some truth, but each one will also vary depending on the situation. The question at least worth thinking through is whether the AI industry itself will continue.
If the subsidies currently sustaining AI disappear and AI model API prices rise, can this situation persist? When no one knows whether current AI pricing is sustainable, how much confidence can anyone place in the future of a labor market being restructured on the assumption of those prices?
By then, non-AI code might actually be valued more highly. Right now it truly is available as cheaply as water. The price has fallen dramatically, and if frontier AI models start rising in price again, will they really be priced below the value of human labor?
I find myself thinking about AI-assisted coding and the game of Go.
Go has fixed rules and a fixed objective, and within those constraints an enormous number of choices unfold.
Programming is similar to some degree. The programming language, the execution environment, and the requirements constrain the available choices, and the programmer finds solutions within them.
In Go, humans spent a long time distinguishing good moves from bad ones and in the process built up joseki. Then AI arrived and revealed that many of the moves humans had long believed to be good were not necessarily the best.
Someday, code we consider beautiful may be evaluated by AI as an inefficient solution. Just as in today's Go world, there is a real possibility that programming conventions will shift to center on the solutions AI proposes.
Seen from this angle, the debate over taste is also worth reconsidering. Taste was described as the last means of differentiation left to humans in the AI era. But if AI begins to reshape human taste itself, how long can that differentiation hold?
Unlike Go, though, programming has no single win condition. There are many cases where even what needs to be built is unclear. And really, how pointless is it to try to predict the future.
In the end, one side is certain of human value based on problems AI cannot yet solve, while the other side is certain of the disappearance of human labor based on problems AI can already solve β but the labor market will not be determined by AI capability alone. Interest rates, capital costs, regulation, and many other complexities will all feed into whatever outcome emerges.
Looking back at it all, every debate in the programming community can be described as a story about how to sell convincingly the "certainty about things that cannot be certain." I think this is the essence of business.
Something I've been noticing lately is that any attempt to connect with others always comes bundled with a transfer of their anxiety onto you. I wonder whether that is really necessary, but I also think that selling imperfect solutions to anxiety might just be what business is.
In the end, to make something and sell it you have to understand other people's anxiety. But being constantly immersed in other people's anxiety feels like it is enlarging my own neurosis.
In that sense, isn't the independent developer always a creature living right up against the edge of madness? Or maybe I am just too critical of everything. That's probably why I'm not very popular.