Clojurians Log v2

In my view it seems all these articles come down to a simple rule. Don’t outsource your brain to a tool.

The disturbing thing here is that, despite his list of how “AI usage generates a fragile codebase”, he is writing 100% of code using LLMs at his new job.

Thought’s been around pre AI though - the turing tarpit https://en.wikipedia.org/wiki/Turing_tarpit

“At some point, AI will do most, if not all, of the coding” is pure speculation, so it rests on an assumption about an unknowable future.

I think even if we accept the proposition as true, we’ll still have to audit the code, and so the choice of programming language matters.

My second thoughts is that whilst languages are Turing complete, they can all do the same, that is in theory. In practice that simply isn’t the case. Taking the argument to the extreme we should always pick assembly language.

of course the language choice matters, even from a token perspective

and obviously like before, you wouldn’t write embedded microcontroller code in ruby (I presume)

So being an architect in a company which uses AI, it does make you wonder. Let’s say someone suggests rewriting the frontend to next.js, and does it in like whatever a day with LLM, and let’s for the sake of argument say that it’s correct and ready to ship. This is somewhat feasible today, but previously it wouldn’t have been, it would have been an effort over months or years, and we’d do a proper due diligence before starting. Now, it’s like if we can rewrite to next.js in a day, the decision to do so becomes a non-decision, because if it doesn’t seem to work, we’ll just rewrite it to Clojurescript the next day. So, like what does one then care about as an architect?

you better have a stellar test suite, I don’t think porting to another framework results in bug-for-bug level compatibility…also why would you need to rewrite if language/framework doesn’t matter?

why rewrite in Rust if llm can just as well write C?

Ruby does have a implementations for running on microcontroller, but you point stands. You wouldn’t write a database in Ruby.

most likely the unstated assumption in the argument is that from the set of practical languages for a given type of problem (say a web app managing business processes) then the language don’t matter.

some language are easier to do for certain kind of tasks, whether you are using ai or not, i was coding for a authentication project which used c# and it takes 6 month for 5 engineers , if technology was erlang it would take 1 engineer for one month. So tech-stack plays vital role

I disagree @U0BHKM127SQ (a bit). I don’t think big tech’s main goal is to replace programmers or languages. if anything, looking at their business model, the more there are, the more money they stand to make. Corporates might though, although every study done so far seems to show that there is no efficiency gain. I do think that big tech is looking for more efficiency and quality in code writing. Writing more and better code in less time seems to be their main focus now. As a developer (and I am hardly a developer myself) our goal is to ensure that the tools give us the best possible outcome given time and money constraints. Even though an LLM is quite capable of producing code by itself, I see it as my task to ensure that it writes what I actually want and need. Which means I need to be in control, not the LLM. Perhaps that is overly optimistic or naive, I don’t know. But that is how I see it.

I would really insist on getting an answer to "Why not clojure then?" as a reaction to that question. At the very least, that would force the person making the statement to make the statement more precise - and probably uncover a few contradictions.

Hahaha! I actually tried that once or twice, @aaronrebmann. It kinda’ works. People either concede or start to backpedal and hedge.

Big Tech is definitely trying to reduce their workforce, especially their engineering workforce as it's the most costly one they have. They've been offshoring and now trying to get more done with less using AI pretty intensely. Lots of big tech has moved to a renting model, rent you a subscription, rent you hosting, rent you hosted AI models, rent you GPUs, rent you office suite, rent you a space to publish your videos, publish your photos, chat with friends, etc. It's not some special secret goal, it's all about chasing profits. If growth reappears, and they feel they need to hire a big workforce of engineers again or even pay them more they will. But currently growth has stalled, the only growth driver has been AI which doesn't require much human workforce, you just need data-centers so they're even trading headcount for data-center build out which is one driver of their massive layoffs.

Emphasis on “at some point”. And “we are nowhere near that point”. Right now I spend waaaayyyy too long twiddling my thumbs waiting on AI. If I were not already on TestFlight (thx AI!) I would start over and just use AI for stuff that would have me digging around SO, or perfectly mechanical things AI can handle. So of course I still want CLJ*. And having seen where AI struggles, it is not clear we will ever be able to treat it like a compiler and trust its code.

I agree with @didibus. I see Alexander has CEO in his blurb, but I’m guessing it’s a small company. Mid- and large-sized software company CEOs are saying outright that they want to fire employees.

As for the implementation language LLMs use - I’m quite concerned that languages that focus on developer ergonomics, like Clojure, will not be used. The Java Architects were just saying that “when AI generates the syntax, the runtime becomes more important”. https://www.youtube.com/watch?v=c_8ElIjbstk (around minute 22) In their scenario, code will be read, not so much written. If people aren’t writing Clojure they won’t be able to read it. Many of the great things about Clojure require practice from having spent time writing in it.

I’d maybe re-label your architect archetype as ‘property developer’ 😉 In my experience, a good architect is pretty detail-oriented and might fall somewhere between the two positions you’ve outlined.

I think a lot about what Terry Davis said about temple OS as compared to Linux and windows: that they are “tractor trailer trucks” and that he was building a “motorcycle”. Windows/Linux were building these giant powerhouses that had to contain l constraints and thoughtful consideration of edge cases that Temple OS didn't bother considering because it was not a part of it, or Terry’s increasingly schizophrenic value system. He goes on to say “if he leans on it wrong/too far that it will kill him” but it's fine because he knows what he is doing, dogfooding his own stuff, and not building for someone else. I think the fact that he was making Temple OS for himself (and I guess god?) means that who cares what other people think of TempleOS… Small anecdotal/personal addendum: I similarly make tools for myself first and foremost, personally; and I am only dissatisfied with my code if it doesn't meet my own requirements; which may change based on feedback from smart people who take a look at my code, as I actively look for constructive feedback on my projects… Programming every line manually by hand is a safe and responsible way to code for people who want to work for corporations, generally fine tuning an already extant system which is probably in both of our opinions, overly complected anyways (and often literally doing something evil in a capitalist sense at the end of the day, if you consider recording data about people immoral). I would first suggest that coding with LLMs and other future ai tools is not only not problematic because the argument that it somehow causes brain damage from use is a farce; a reductive take from the same group that once said we shouldn't use Wikipedia to do “serious” research. I don't write research papers quoting Wikipedia but I think it's safe to say that if I want to look up the name of a particular Godzilla film then Wikipedia is fine and I barely care if it's wrong anyways even though no it probably won't be wrong. Also why would we just give up on an idea because one implementation of ai (namely, letting it think for you), is bad. Driving my car into oncoming traffic is bad but I don't do it because I prefer getting where I'm going and being alive. Secondly, and harkening back to the main point I am trying to outline: sure I don't think a bank should rely on my shitty localhost:8888 web app that I asked grok to make over a weekend for 200 dollars (which I've never done and would never do (because why the hell would I ever use grok?)); but that strawsituation is not a real thing in the first place (IMO); and most applications made by true software developers working on these technologies obviously implement refined techniques and procedures to ensure published code has some veracity; and would otherwise responsibly mark it as possibly broken or in an incomplete state or something else. This was a lot but one last note about the proposed “automation” vs “tool” note is that it applies to “ai as coding agent” but of course ai might be implemented as a tool in the app for god knows what (so many possible uses), which an ai might program itself; but is a separate use case not recognized often. There are also of course so many other ways than as a coding agent, some of which may be more or less garbage than others and I think it would have to be considered on a case by case basis.

Shouldn't you flip the words? Builders are more concerned with getting the house built fast, cheap and flipped quickly and on to the next one. Architects like to take their time, make sure every detail is perfect and fits the overall vision, etc.

I think I agree with the crux of this argument, and have thought a lot about this blog post since LLMs took off, I think it goes a long way to explaining the differences in peoples reactions: https://josephg.com/blog/3-tribes/

you may find it interesting to read Alan Kay (he is mentioned in the article above) opinion on llms • https://www.quora.com/Can-an-AI-command-line-replace-the-GUI-as-the-primary-user-experience-for-computers-assuming-the-technology-improves-and-irrespective-of-today-s-state/answer/Alan-Kay-11?comment_id=511419029&comment_type=2 • https://www.quora.com/What-does-Alan-Kay-think-about-programming-and-teaching-programming-with-copilots-and-LLMs-of-today