Clojurians Log v2

How in an IT company no one knows SQL?

This is the truth I learned over the course of my career. For whatever effing reason, it is believed or assumed that every programmer knows/must-know Javascript, Python, SQL and Bash. Reality? Almost nobody knows.

To be fair, whenever I start thinking that I kinda know SQL now, something else invariably comes up and bites me (not in the “pure SQL” sense though - just the particular PostgreSQL impl of SQL).

AI, but pretend there is no A!

> (not in the “pure SQL” sense though - just the particular PostgreSQL impl of SQL). @p-himik What’s “pure SQL”? As far as I know, no actual implementation implements, say, exactly (and only) ANSI SQL. In my experience, invariably you’re going to need to deal with the details of a particular implementation if you’re doing anything more than the most basic of queries.

developpers that don’t know a thing about SQL and don’t seem to be using even a simple harness, I don’t see how that’s AI skepticism, more like human-skeptic here. I bet their pre-llm code is full of ORM abuse and queries that should take ms are past 1s and they use all kind of insane workarounds that wouldn’t exist in the first place if they had a clue how their ORM works underneath?

This is a surprisingly good way to learn SQL… https://exercism.org/tracks/sqlite

or the older https://use-the-index-luke.com/

> developpers that don’t know a thing about SQL and don’t seem to be using even a simple harness, I don’t see how that’s AI skepticism, more like human-skeptic here. @yenda1 One of the arguments against the use of LLMs and agents is that it is making the development experience worse in many aspects. I think @mauricio.szabo is pointing out a case where developers are leaning on LLMs to solve problems without understanding them. And part of this is also because use of LLMs and agents is being encouraged by management.

I think you’re right that this is due to how people are using LLMs/agents and that they could use them better (and reflects on them as developers independently of their use of LLMs/agents). I also think that this kind of behavior is, as you point out, human (or at least that’s how I interpret your use of “human-skeptic”).

Yes, and also, let’s consider that nobody checked two very obvious “weird” conditions; then, when I asked the LLM why these conditions were there, it literally admitted “I had a list of example IDs that this query needed to return, so I cheated the results”

There’s a tendency I see on the “pro” LLM/agent side (or at least the “anti anti LLM/agent skeptic” side that will only consider arguments that are intrinsic or novel to LLMs/agents. I think ignores the human context in which LLMs/agents are being used. That some of the issues LLMs/agents contribute to are not new doesn’t mean they can be dismissed. In many cases what’s new is that LLMs/agents are making these existing issues worse, and sometimes disproportionately so. The idea that “well, if you just use LLMs/agents in a more ideal way these aren’t going to be an issue” ignores what’s happening right now and the direction it’s currently going.

Again, that’s not to say there are no benefits to LLMs/agents in software development. I think those need to be weighed against the downsides, and not only the downsides in software development.

Completely agree - one problem I see is that, for example, the company is pushing hard for LLM usage; then I though “shouldn’t we talk about this before”, and then I remembered that all our channels have some kind of AI on it - Meet, we have Gemini summarizing discussions, Slack, the bot summarizes channels and huddles. With this, anyone talking “maybe we should review our AI usage” is essentially “one search away” from being outed as a possible layoff option so… nobody says anything.

> from being outed as a possible layoff option so… nobody says anything. Yup. It’s really scary. 😔

I also tested here, on my side, on asking a LLM to search our Clojurians slack to check blog posts of people that speak good or bad of Clojure or one of its dialects. I didn’t actually connect the Slack because it was just a test… and even then, it still brought me some very interesting posts. So imagine a company adding a scheduled procedure searching all company’s developers’ blogs, trying to match which ones are skeptical…

trying to match which ones are skeptical…

> “I had a list of example IDs that this query needed to return, so I cheated the results” Which model/version of claude is it? because this ass-licking behavior when caught doing mistakes was more prevalent in previous models, and unless your claude is somehow sharing conversations between all users or it signed a commit or something, it cannot possibly know that it made this mistake, it’s just being a people pleaser

@yenda1 Opus 5.5, High effort.

claude code / desktop or web?

well if that was some source file under git it certainly had its signature on it

I would recommend you to push for some proper software engineering, hardcoded IDs wouldn’t go through a simple write test / code / review loop with different agents. A lot of best practice for regular development are still valuable with AI. One-shoting photoshop only works on twitter and reddit 😄

> Completely agree - one problem I see is that, for example, the company is pushing hard for LLM usage; then I though “shouldn’t we talk about this before”, and then I remembered that all our channels have some kind of AI on it - Meet, we have Gemini summarizing discussions, Slack, the bot summarizes channels and huddles. With this, anyone talking “maybe we should review our AI usage” is essentially “one search away” from being outed as a possible layoff option so… nobody says anything. That’s kinda wierd, there’s a whole spectrum of what “review our AI usage would mean”. It can go from not use it at all to double down in one dimension, and keep doing things but do more of it to how can we improve how we use it”. The question is what is the strategy the company as decided on and regardless of what you say in meets and slack, if you’d rather handcode in a company that is betting on 0 human written code its in both parties interest to end the relationship.

> if you’d rather handcode @yenda1 What’s your opinion of people who would take this perspective?

If you tell me “I want to write clojure myself” I’ll tell you that’s what weekends are for 😄

@yenda1 What kind of world do you want to live in? A human-centered one, or one where humans are resources, and they can do human stuff on the weekend? Or do you see an alternative to those two?

We’ve been living in a world where humans are resources and can do human stuff on the weekend for a long time now. IT people just happened to be a very valuable resource for most of that time

I can’t tell about the whole world but I feel like we are way more human-centered in my company now that the bots and agents are all around. It’s like we’re all management in a big software factory with tons of workers that happen to not be humans

Agreed. And again, is that a world you want to live in? Or just accept as the world as it is? These are not fixed rules, like the Plank constant. These are societal rules of a game we’ve been playing. We can change the rules and the game, maybe even just a little. And the fact that it’s hitting IT now does mean it’s hitting us software developers now.

I’m glad you have a place at your company that’s human-centered. That’s a good thing. If you look around, I think you can admit that that’s not the world as a whole, or even just the software development world we live in.

yes that’s what I’m saying, I’ve been in the development world for a while and it’s never been human-centered. Even during hiring booms and crazy salaries periods it was always just about offer and demand. the hit started before AI from my perspective, when the unlimited VC money dried up around 2020. I’ve actually not seen much of that VC money for AI in Europe, it’s been much more conservative there. At most investors request it as a must to ensure you stay competitive

@U0BGC50RM26 indispensable until management believes otherwise. And how will anyone really learn DB basics or AI basics or anything else going forward?

Why were they asking you to “check” it? Too slow? Too fast? Incorrect results?

> Why were they asking you to “check” it? Slop bomb, essentially…