(Following thread was prompted by people pointing out that the Bluesky dev team seems heavily into vibe-coding now and originally posted on said vibe-coded Bluesky platform that is now constantly failing.)

Over the past year, every single time one of the apps or services I use suddenly became less reliable and more buggy, I never have to look far for the "Claude is amazing and now writes most of my code" post for the devs involved.

Best part? It's always somebody with years of experience. Exactly the demographic that is supposedly able to use this shit safely, but my impression is they're just as bad as the novices

This is happening IMO because of one of the fundamental issues with software dev (and this predates "AI" and was one of the themes of my first book):

Most software projects fail and most of what gets shipped doesn't work. The way the industry is set up means there is little downside to shipping broken software

Few devs have a reference point for genuinely working software. Usability labs were disbanded over 20 years ago. Very few companies do actual user research, so their designs are based on fiction. Bugs are the norm

Alienation is also the norm for devs, both socially and organisationally. Whether it works for the end user doesn't cross their mind. Whether the design fulfils business needs is not their problem. Bugs are a future problem. Ship insecure software and patch it as user data gets stolen

Devs are so disconnected from the output of their work that many of the norms of the industry are outright illegal: there's a good chance that if you follow popular practices for a React project, for example, you'll end up with a site or product that violates accessibility law in several countries

Few devs would even know where to begin to look to answer the question "does my software work for the people forced to use it?"

Because the element of coercion and a complete disregard for consent is now an integral part of how the industry works, but that's a topic for another day.

@baldur Whoever came up with 'Yes/Not now' needs to be dragged into the streets and shot.

No wonder some folks don't understand consent - our software doesn't allow for it.

or the "You must allow our JavaScript programs to run on your browser, otherwise we won't allow you to get to the information that we're legally required to provide you with"

CC: @[email protected]
@lxo @baldur Don't forget the 'ole 'our site runs better in our app (and we won't let you view our site without it)'
@DarkestKale @lxo @baldur and then the app redirects back to the web browser for authentication anyway.
@baldur I do wonder how much of the disregard of consent is an artifact of being owned by the Epstein Class and the wannabes making this seem normal and grown up
@otfrom Yeah. I've been wondering the same.

@otfrom I think it’s a mistake to blame it on Epstein. The consent / coercion balance change arises from licensing vs ownership, and the extractive nature of sociopathic billionaires.

A billionaire is going to billionaire, (or a baron), because they’re sure they know best, and other people’s rules don’t need to apply to them.

Sure, a whole chunk of them will want particular expressions of transgressive control of the sort Epstein provided, but the control and coercion were there already.

@hypostase that's why I said Epstein Class

The control and coercion are what they are after. The material wealth (billionaires) is one way to get it (there are others, but money works well)

@otfrom Fair. I merely claim that an Epstein Class is a subset, and that, as popular a bugbear as the label is now, people like Ellison and Gates have always been bigger contributors to the problem. And that’s leaving aside the South African mob.

@hypostase I see Epstein as a synecdoche for the whole worldview. I don't mean any particular individuals or scene.

They revel in being able to trick others. They think they have the most merit. They think they can take what they want and deserve it. They think our lives don't matter (or at least not as much as theirs)

@hypostase though if any other name worked to properly get across what a problem the oligarchs are then I'd use it

@otfrom whereas I see Epstein as a lesser player, who crossed a very particular line, and was then being hung out to distract from the fundamental behaviour of the others. Wyrmtongue more than Saruman.

What he did was horrific, and needs to be stopped, and was hiding in plain sight, like ever so much child abuse.

But, much of the danger of the oligarchy, and their process of control, is not only in plain sight, but considered a necessary part of capitalist structure, and therefore good.

If we are to stop the former, we also need to stop the latter, and that’s harder to do without acknowledging that it’s there.

@hypostase ah, that might be where we differ then. The pedophilia was just part of Epstein circle. Taking children, taking money, threatening governments via insider knowledge and market collusion -- it is all there.

All the taking, and the only limit for them is where their whim takes them. They live to take what they want when they want in all things and they don't care about who gets hurt. Their feeling of supremacy protects them.

(Or at least this is what I'm seeing in the Epstein, Mandelson, Prince Andrew, Steve Bannon, Richard Branson, Elon Musk et al emails at the DOJ)

@otfrom yep. It’s all there. The danger is in thinking that it’s the only place that sort of collusion happens.

US gerrymandering, for example, is not fundamentally Epsteinian. It does introduce a very particular form of political corruption and control, that really needs to be addressed if consent is to become a forward part of political thinking.

But that’s not new, either.

@baldur This is the first convincing explanation I've seen for why software engineers would not object to these broken tools.

@baldur years of industry "leaders” with new design patterns, a dozen development methodologies with the (sometimes misguided) intention of doing better software. Went from jokes about places using LOC as a productivity metric to bragging about how much (and only) code can be output

Deeply unserious industry

@spinnyspinlock @baldur You could easily be describing the entire culture of capitalist management since WW2. Endless people management “strategies” that are just elaborate rain dances to avoid saying “pay good money, hire enough people to do the work, and treat them all with respect”.
@naptowncode @baldur It was not what I meant (valid interpretation though), some people do legitimately want to write good software. Or so I thought
@spinnyspinlock @baldur Oh I agree! I think most devs want to write good code (and most managers want to manage well) but we’re all human and everybody wants a shortcut around the methodical work that is easy to describe but hard to do consistently well. Plus there’s bandwagon effect, learned helplessness, and a million other psychological traps that drag us away from the plain “just eat your damn vegetables already” style of work.

@baldur really interesting stuff I've not seen much before, thanks.

Can see how this stuff is enticing to someone just starting out but it blows my mind that experienced devs want to use it. Turned off several long time reads/listens because all they talk about now is AI

@baldur The disconnect (or alienation) is so baked into the industry that you'll get flagged as "potentially problematic" if you express field experience in the industry the software you interview for working on is aimed at.

@alda @baldur In my previous workplace, e-health, I pointed out that certain corner-cutting would cause patients to die.

My team lead: "You should come to the office to get COVID, so that you think less."

@rhelune @baldur Yup. People like that are so used to business software being shit that even if healthcare should be considered life critical, stuff like this isn't considered to be "real engineering" anyway.
@alda @baldur It seems to me that the field (management positions in any case) is attracting people without conscience.
@rhelune @baldur The game is about hiding your incompetence while pleasing those above you long enough to get promoted.
@alda @rhelune @baldur oh, so much this. I would not have wanted to play that game even if I could, but not realizing the others around me were playing it really fucked me over.
@crowbriarhexe @rhelune @baldur This is basically the first thing that comes to my mind when I hear or read something akin to "autistic people have a difficult time getting promoted in the workplace".
@alda @baldur a friend of mine said "EAs exist so programmers don't have to talk to BAs, and BAs exist so that EAs don't have to talk to business representatives" which hit me like an arrow

@zanchey @baldur "Look! I already told you! I deal with the goddamn customers so the engineers don't have to! I have people skills! I am good at dealing with people! Can't you understand that!? What the hell is wrong with you people!?"

https://www.youtube.com/watch?v=fcIMIyQnOso

A lot of business relationship managers and team leads got into their positions by hiding their incompetence whilst pleasing upper management long enough to get promoted.

Office Space - "I Have People Skills!"

YouTube
@baldur This reads like an open Work–Feedback Loop: work ships, but real user effects don’t reliably change decisions or next work. Maybe that is a good starting point to discuss this issue in the organization. Thats at least what my thinking model is for :)
@nobsagile @baldur And when development stops because the rot permeated the tools used for development the leadership says "Oh, something broke. Let us think about it". That is unless they consider declaring bankruptcy the easier option ​
@bunny @baldur Yep. “Let’s think about it” is the sound of years of postponed fixes finally cashing out.
@baldur I think alot of it stems from the “move fast and break things” attitude under VC pressure, that now includes breaking their own software. “Agile” is to blame as well, which tends to prioritise siloed rapid feature development, all too often these are hacks to get it working within the timeframe allotted. This ignores broader impacts in the codebase leading to rapid bit rot.
@baldur Disillusionment is the name of the game. :-(

@baldur I still remember, talking to a twitter dev who had an utterly ridiculously foolish take on XYZ issue go viral.

They told me 'Uhhh, I've never had this much attention on me, my tweets never go beyond my social circle. I had to turn off my phone. It kept buzzing.'

... this was a person who worked on the UI. No shit they had no idea how to deal with high volume 'oh, you just 200k likes' kinda shit, they never experienced it themselves.

@baldur You're dead right. So many devs are pathologically isolated from the people that... use their stuff.

@baldur …and so let's have software-defined vehicles. Self-driving ones! Full Self Driving™ ones!

Putting wheels under it and setting it loose in traffic prevents any faults and nullifies all bugs in software. I mean, that's just science.

@baldur It's not true that usability labs were disbanded. They just shifted targets. From usable software to engaging software. Often engaging to the point of getting into a legal grey zone. OK, maybe they do not deserve the name usability labs anymore ​

@baldur Honestly I think a big part of it is more than our industry being deeply immature still; I think the most important throughline of the research on LLMs' effects on cognition is a consistent attack on metacognition, which seemingly doesn't abate with experience. The same corrosion happens to juniors and seniors alike, but the seniors have more rationalizations at hand to pretend it doesn't.

(Speaking of, that "cognitive surrender" paper is the latest in that theme: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6097646)

@delta_vee ooh that's an interesting angle (paper)
@dngrs Oh yeah, it's a good approach, and IMO has useful explanatory power for some key things we're seeing

@baldur Apropos of nothing, the absolute worst implementation of Raft I've ever seen in my Raft course was by a pair of senior devs with a combined 60+ years of experience who decided to pair program together and announced ahead of time to the group that they were going to "win" Raft. They did not.

An undergraduate who'd never coded with sockets before did reasonably okay.

@baldur motivated reasoning is one hell of a drug. I've seen a developer far better than me getting completely hypnotized by LLM sycophancy; I tried pointing out that what they proudly posted as "see? Completely bug free after just a few rounds of conversation!" did in fact contain a subtle bug/violation of their prompt. Got ignored and they only went downhill from there. It's a saddening cult really.
@baldur Real life experiences in this are interesting to hear. What I note is that I can make/fix my own code with help of AI because what is "shipped" is often not what I want, or broken... and the blast radius of my code errors is limited because no-one uses it but me.
@baldur Some of the most defensive people I've ever met were mid-career software developers. They quite obviously tried to protect their fragile egos (and possibly careers) from valid criticism directed not at them but at their software. I have no doubt in my mind that those are exactly the type of guys who were immediately hooked on LLMs.

@baldur I feel like having spent most of my career building embedded systems aimed at industry rather than consumers, where customer support issues can mean sending a technician out with a USB stick on a ten-hour road trip, has insulated me from the worst madness.

If your sloppy coding breaks a manufacturing line or distribution network, bugs become expensive fast.

Though having said that, $CURRENT_EMPLOYER is pushing for greater use of LLMs in our workflow...

@cunobaros
You could use it to respond to all emails so you can focus on properly submitted tickets and not meetings. Ya know if they're going to force the issue.
Just a thought.
@baldur

@baldur I suspect that these people are so fixated on automation that they’ve moved to agentic generation of their code within days of adopting the LLM.

I probably babysit and check the LLM’s answers 1000% more than most people and it STILL attempts to introduce absolute junk into my code and makes me look like an idiot, reassuring me that my (actually stupid) idea is brilliant time and again.

Case in point: Claude Code runs in THE TERMINAL on a custom **react game engine**. 🤦🏻‍♂️

@baldur This has been on my mind the last few days, too: https://mas.to/@nielsa/116171030173125331

I see so many people falling into LLM delusion, who I thought would know better, with no seeming pattern in *why* they fall for it.

Yes, the lack of negative incentives is certainly a factor.

My best explanation so far is that LLMs are kind of "acting" like 17 cons (some new, some old) in a trench coat, and different combinations of these trick different people who'd be able to resist most of these on their own.

Niels Abildgaard (@[email protected])

@[email protected] Yeahhh I've been thinking about this type of thing recently, too https://mas.to/@nielsa/116155283958385166 It's an eerie, lonely feeling to look at those one thought of as good at the exact things needed to see through the flashy deceptions of LLMs not just fall for them, but commit fully and strongly. And, in my experience, then refuse to face factual points that should convince them to let go of the delusion...

mas.to
@nielsa @baldur I've been thinking the same. LLMs seem to be incredibly good at degrading cognitive performance in both extreme and subtle ways. It's partly why I stay the heck away from them
@nielsa @baldur Some of the scams are nicely described here: https://softwarecrisis.dev/letters/llmentalist/

The other part is that the IT slop industry has perfected ways to make sites and platforms addictive to retain users, and all that experience can and likely is funneled into new platforms, such as AIaaS. Some people clearly fell exactly for that
https://www.linkedin.com/pulse/just-one-more-prompt-patrick-patterson-fhsqf/
The LLMentalist Effect: how chat-based Large Language Models rep…

The new era of tech seems to be built on superstitious behaviour

Out of the Software Crisis

@baldur

Most software projects fail, most of what gets shipped doesn't work, *and also* most of them never needed to be written in the first place.

It's easy to ship a product when the purpose of the product is something like "discourage people from using medical care they've paid for."

@baldur my ongoing theory is that this is also an example of vibe coding appealing to the absolutely worst amongst us who also manage to gain high "status" within the industry - it's GREAT if you're someone who subscribes to "promotion-driven development"