They generated lots of documentation across the whole stack and now makes all PO/BAs read it if it's correct. So not just 300 lines - he unironically generated thousands of lines of "documentation" and is now making hundreds of people review it for him
Complete brainrot
Au psychosis is getting seriously outrageous at this point
Thankfully I'm a dev and thus aren't in the blast radius of that genius idea
I see no difference there except speed at the cost of whatever little understanding may have been gained by the manual inspection between steps.
Personally I've found LLM's suck at multi-threaded applications. (Because I've been tempted by the ~agentic loop~ and been burned. Then I hand code the core logic and all is well).
Woe to the developer who tries to prompt their way through this.
But seriously, lots of apps are just glorified NextJS apps which have tons of training data. Something like rust would likely churn out nonsense that compiles eventually but isn't optimal.
Agreed with the caveat that I think if you know what you're doing and are very cognizant, LLM generated Rust is amazing. I feel like it's hard compilation requirements gives a guardrails for a LLM and if it compiles, you're pretty safe against memory issues.
x=ctrl-c("output error report")
eval(ctrl-v(x))
sleep(10)
user("Input...")
You're basically describing my work day. My job went from interesting, highly skilled work to mostly horrible brain rot over the past year.
It's depressing and I don't know a way out. Other then a complete career change, throwing away 30 years of experience and a matching salary.
I only have 10 years of experience, so I'm not trying to say that your lived experience is invalid, but personally I figure if this is the way the industry is heading then I may as well try to learn how to thrive within the new environment.
The current hype cycle might even represent a net global gain like some people argue. But it represents yet more externality driven exploitation. Remains to be seen what the new equilibrium will be given that now the population being exploited is not only close to the core of the world system but also already overburdened.
I do think it represents a mere acceleration of the previous trend that resulted since around 2005 in an explosion of average (not median) pay for software developers due to similar dynamics. If the system settles on a new level of pay that manages to convince enough people of enough skill to go along, it might just hum along and not implode. As for the rest of us, welcome to the growing permanent underclass and brace for the impact of the climate wars. We will be the fodder that will insulate the chosen ones under their air conditioned domes.
If you’re unhappy, make a change.
The real question isn't if things are going to change around our jobs, its are we going to be able to move fast enough to avoid starving in the streets?
But that's because I don't have any real familiarity with the systems involved and I don't expect that gaining such familiarity will benefit me. If I am working on a system or product I'm responsible for at my job, it should be a different situation.
But I think we should decouple mediocrity from laziness. I haven't seen any team invest in the mentorship required to develop juniors in years, for example.
It would make sense to develop juniors if most of their comp was a four year vest but that doesn't happen until later. And in your first year or two you're usually a net negative... This is even more true with AI.
Especially if you get a little burned out and can't bring yourself to contribute to a side project, but still want to do programming-ish things that get your brain moving
Maybe this «spearheading AI person» just sucks at AI related stuff, as clearly that approach is bananas, but they could still be a OK developer.
If working with LLMs effectively means accepting subpar results or be a reverse centaur, then I’d be glad not to be able to work with them.
I’ve never seen a good example where AI is a net positive to any development workflow. No one argues against compilers, build tools, IDEs, task runners, deploy and orchestration tools. Because they are great levers that lets you create more with less effort.
Where it shines is glueing systems together or building one-off automations that would take days, or weeks, to figure out. It’s for things you don’t have time to figure out or didn’t think were possible.
Why would it mean that? That's one way of using them, sure. Personally, my code is better as I have more time to think about the software design than before, and I'm less avoidant of refactoring in my personal projects.
> I’ve never seen a good example where AI is a net positive to any development workflow
Alright, does that mean you also believe it's impossible then that anyone out there is using AI in a "net positive" way for their development workflow? Or just that you've never seen it, but you're open to it existing?
This one. Only a sith deals in absolute.
I don’t mind experiments to try to find methodologies for those tools. And I believe there are instances where they’ve been successfully used. The issue I have is the kind of generic statements that they are good enough to replace currently established methodologies. Like using AI is a panacea.
> Personally, my code is better as I have more time to think about the software design than before, and I'm less avoidant of refactoring in my personal projects
That’s a bit what I’m talking about. Have you investigated how it has helped you? And if there are other, more economical way to get the same result? Your statement seems more ritualistic than logical.
No I haven't, but I'm happy to just freeform walk you through my thinking on it: I typically write (wrote?) software for two purposes: consulting/freelancing for others so building what others want, or for simplifying and making my own life easier and more enjoyable. "Stupid" stuff like Home Assistant for example, isn't really life-or-death, or Jellyfin for that matter, both things my family relies on now, but our daily life just gets easier all throughout the day when everything works in sync with what we're doing.
It used to be I had to make a decision what to spend time on, either I work on my professional stuff so we have enough money to survive (maybe more) and I get new challenges and all that, or I spend time improving and maintaining my home infrastructure, or whatever software I feel like I'd need to be better at doing my professional development.
I no longer am making that choice, I'm spending less time in front of the computer, yet the output and quality of my work remains the same, and the code and design when I look at it, even stuff I shipped 6 months ago, I'm still happy with how the code is, which for me I guess is the way I validate if what I produce is good enough.
Nowadays, my entire home-lab is configured with Nix and almost everything except my workstation and some random stuff, runs NixOS. Everything is hosted on a local Forgejo instance, which also has it's own (custom "written" of course) agent acting on issues and PRs, and I have my harness basically maintain my entire home lab at this point. Now I just open issues, have a conversation until everything is 100% clear, end up with a PR to review and merge if it looks good, and I can do this while juggling other things.
I agree with you that there are tons of people who are selling LLMs as a panacea to lots of things, and there is so much over-hype in the industry and ecosystem, I also feel like every "new thing" kind of comes with this type of almost scamming, which sucks, and makes it hard to discern from real positive opinions vs just regurgitated opinions someone read somewhere. I'm not sure what the answer to that is, except perhaps as what you say, only a sith deals in absolutes.
What are the exact "currently established methodologies" you're talking about that cannot be replaced by LLMs + a harness today, just as some examples? You're probably right that those exists, but I'm curious to hear what you think would be the most difficult to replace today.
I was explaining [0] under another post that programming is mostly translation works. You take a specs and you formalize it using code, like going from sketch to a proper engineering drawing. Software design is more creative, where you take a problem and then comes up with a solution (creating the specs). Software Engineering is ensuring that those two are done well enough while consuming the least resources.
So a program is always a formal system. It's also static. It will be executed by a computer which will actually have a tangible effect in the real world. That effect is what's valuable. The program is the seed which let us control that effect. Aka it's the map that let us plan the journey, but it's not the territory that we will have to travel in.
The issue I keep pointing in most of my comment is thinking that the map is the territory. That the novel are the words and not the story so we need more words. Or that the code is more important than the user' workflows, se we are adding more buggy code, while not ensuring that the workflows are undisturbed.
> Nowadays, my entire home-lab is configured with Nix and almost everything except my workstation and some random stuff, runs NixOS. Everything is hosted on a local Forgejo instance, which also has it's own (custom "written" of course) agent acting on issues and PRs, and I have my harness basically maintain my entire home lab at this point.
It's also highlighted here where you focus more on the process than the output here. The goal is to have a working homelab. NixOS managing it is only the process (accidental complexity). If it's where truly about the goal and not NixOS and using AI, by this point, adding new nodes (software, devices,...) should be as easy as selecting it and adding it to the current system, like a strategy game.
You can see that philosophy in OpenBSD, where the focus is to have a working OS, not to work on developing an OS. A lot of software are done and it's mostly just bug fixing every once in a while. You can also see the same attitude in industrial engineering where you develop a solution and then use it for years. You don't spend all your time tweaking it and thus disturbing the production flow.
So yes, when I see a LLM methodology, it's mostly about the work itself, not the output of the work. There is no definition of done or even the idea of having one. It's work for the purpose of working.
Both of these can be quite invisible. But the benefit is there.
Uhm ... I've heard quite alot arguing against IDEs. And I feel IDEs are too much of a lock in.
Over the past year maybe 1.5x to 2.0x for me. As in: I can work on two projects at the same time with reduced amount of context switch. But that's it for me.
Maybe I don't have the brains for 1000x terminal agent coding, but 2 parallel projects seems like my saturation point.
But most who keep on going about their 10x productivity gain are indistinguishable from that one obnoxious guy at a party who won't shut up about his Ayahuasca retreat last spring. And they think they're Erdős.
AI is like having a junior dev with an adderall addiction and an encyclopedic knowledge of coding syntax at your beck and call.
Many many people have jobs where their contribution is granting access to deliberately undocumented things, like knowing where the config files are and some such. They hate the idea of AI. For my non IT friends its great for diagnosing wifi issues. It's also great for competent network engineers. It's not great for those who gain a salary due to having memorized some actions or settings that they don't even understand much. Note that this group has also already resisted traditional script automation, just like US dock workers who resist automation.
Workplaces are where many of us spend a large percentage of our waking lives. So, it is "normal," unfortunately. This is why HR language and norms have broken containment into the non-work world. (Whether that's a good thing is another question.)
There is not one single discourse norm to rule them all and calling out every little thing as too offensive is just annoying. Yeah, yeah there are no midwits, everyone is a unique little genius flower, yeah.
More productive? Sure, to an extent.
Being merely more productive is a tame claim for what some people unashamedly post here
Karpathy, Carmack, Terence Tao, Simon Willison etc. are all smart people and manage to use AI effectively and productively because it doesn't hurt their ego.
You can see my comment history.
I always put myself in the shoes of the reader when relaying Claudeslop. I either use it as a reference and write it by hand or make Claude edit it 10+ times until it's conveying only the needed information, and in simple, clear words, and without annoying Claudeisms.
i just like dont read peoples md files they send me from claude anymore
I would say to him that if he wants me to do his job, he needs to pay me his salary.
Sounds more like job security to me.
Any resistance to this is met with a "But I'm also using AI as the board demands!"
That sounds far preferable to individuals independently asking questions that are not covered by the documentation and independently reviewing a bunch of potentially hallucinated slop that no one else will ever read.
Maybe a better way of doing this is to ask the experts to do the work? But then again, someone who does not understand the subject could ask better or at least different and more approachable questions.
I guess one reason I object to this is asymmetry - it takes someone a minute to take an hour of my life with no back pressure mechanism. Previously asking a question required effort and signified commitment.
So... back to a currency backed by gold?
In all seriousness, actual AI psychosis is my 3rd risk behind 1. public interactive agents and 2. whatever devs are doing with agents to prod
I don't usually like to gatekeep but knowing that "Au==gold" is one of those basic facts that everybody on a forum like hacker news should know and find unremarkable.
You might even say it's puts on sunglasses periodic table stakes.
Actual intelligence might be[0] to know what typos (of any kind) are fun to riff off.
Maybe this didn't hit the mark for you, but it got a chuckle out of me (especially considering the last hype before AI was crypto).
-----
[0] I don't even try to figure that out anymore, in a few years I'll pick the definition that I think best captures my thoughts
See that way too often in comments like this. Anyway, it was pretty likely they knew it was a typo, it was just a funny response.
Too bad there's no real accountability.
yes you are
"The amount of energy needed to refute bullshit is an order of magnitude bigger than that needed to produce it.""
Until you put in at least the same amount of energy required to verify the AI output as you are asking others to put in to verify it, the AI output you are wielding must be considered bullshit. Really, it should be based upon the verification requestor's historical trust score and how many human verifiers are looped in. A Verification Tax, so to speak:
The energy the creator must spend (Ec) is determined by a function of the number of reviewers requested (N) and the historical trust score of that creator's AI outputs (T).
Ec = N * (1 - T)
For high stakes debugging efforts I'll dig in with Claude, have it do a bunch of testing and give me a writeup, and then review it to extract any useful findings. When discussing with other people, I give them the 2 sentences I'm confident in and then link them to the giant doc so they can review it with their Claude.
IMO once you're working with agents, your whole job is context management. I have a little web server running my Claude markdown docs. When someone sends me a Claude snippet with some partial information I just have them prompt their bot to upload a thorough context dump so I actually know what they did. If someone has had an agent look at a problem they haven't thought about yet, I'd rather just get the info from the bot directly.
Consider just using Claude sessions. Claude has hooks you can use to copy local sessions in the repo.
Git LFS can be used to store directly in repo, helping not just with repo size but also with security/privacy/secrets-leaking
Like, I can see getting a strong warning the first time somebody notices you do this, then if you persist you’d have to be let go.
Just like anybody else who is clearly not doing their job.
Why would you tolerate this sort of behavior at your company?
To avoid rushed or incorrect decisions, conflicts intentionally move through gradual escalation. If someone behaves unprofessionally, for example by submitting a code review they do not understand, the first step is to establish the likely cause.
They may ultimately be at fault, but that does not absolve the surrounding environment. The culture may be toxic, deadlines unrealistic, or communication poor.
The best course of action for someone directly affected by uncooperative coworkers is to avoid assuming ill intent. Get them on a call and let them explain their pull request, however trivial the issue may seem. Even if the gesture is misinterpreted, you still have a far stronger position than righteous indignation.
I once worked in a place where a guy would pick on me in code reviews, he wasn't really higher position, but had been with the company longer than I was.
He was just an asshole asserting his dominance and needed to be taken a notch down by a manager and it turned out he didn't understood (or more likely didn't try to understand) the PR when confronted.
If they wanted an AI answer they could have asked it themself.
Do they ensure that it didn’t accidentally overwrite something it shouldn’t have?
Or do they just throw up a PR and assume their little refactoring bash command works fine, then expect their coworkers to debug it for them?
Who goes when companies need to downsize? Rarely executives; they'll always find a way to be retained...hell, they're the ones with the power to decide who goes.
An Ops person who deals with company code base but is not a SWE in the code base regularly has to work around issues with this software (They are OPs). However they have read access to the source.
In the past, they see a bug and open a jira with how it was found for the dev team, and it may or may not ever get looked at.
Now ops can point claude to the repo and it can most likely [but not always] find the issue.
So this ops person can do most of the work of reproducing the bug and finding the problem/solution but doesn't understand the code. Should they not produce the findings for someone else to read/validate? Should they go back and just open the ticket the old way?
No, they should not. They should not waste others time with output that they themselves do not fully understand. That is not even remotely helpful. The other party can use claude just as easily as they can. Using claude does not add any value in and of itself.
"Hi claude said this thing that I don't really understand please look at it for me" is never useful.
> Should they go back and just open the ticket the old way?
Yes. They should do their damn job while respecting the time of the other people they interact with who have their own different jobs to do.
So write to a colleague asking to finish what your cousin/another employee thinks needs to be done. I think most people would say describe the problem clearly so that I can solve it and be responsible for the fix or let your cousin/other employee finish the job and own the solution.
This is offloading of accountability while denying credit.
I'm not sure how to reply to that.
Your example reveals a common organization smell: work not going through the normal process for prioritization and planning. If you've done that, and the work is deprioritized, that's because the company has decided that your thing isn't as important for engineering to do. Get over it and be patient, or ask a mentor for help selling the business on the value of your thing. Creating a PR bypasses that process and burdens engineering with more work (having to understand and rework slop) than if engineering had coded the thing to begin with.
tl;dr: if a company is internally strategically unaligned, organizational dysfunction is bound to follow
By providing a bad “AI” slop solution that you can’t evaluate, you’re actually creating more work for the people responsible for doing the thing. Now they have to understand the slop you’ve given them, explain why it’s wrong/won’t fit in the existing architecture/has other issue you can’t see without deep context. Or, they have to play politics to ignore your slop without angering you and creating more problems down the line. Either way, it’s harder than just letting them do their job.
If I get a request like this on a day when I don't feel sufficiently quixotic, I say sure and just put it through an LLM myself - if they don't bother to re-read it, why should I? This does not have a great promise for a future of work, of course. Also, in my teams, I explicitly tell people not to do that.
This is the kind of thing that spreads. If you have to deal with it all the time you're going to start dissociating out of necessity.
Deciphering error messages can be a bit of an arcane art, at times impenetrable to all except the authors of a system. Understanding the patterns exposed in such messages and, thereby, deriving their meaning is not always clear even to experienced developers.
Engineers reaching out to you (humans) to explain the output of an LLM haven't fully internalized how to interact with LLMs. When an LLM's output is beyond one's immediate grasp, one can _ask the LLM_ to further break down its output, to explain the concepts and symbols that inform its response.
Personally, I have asked LLMs to draw charts and comprehensively simplify their output, often providing cognitive signposts that "I'm confused" about some aspect of what its reporting.
In other words, some people, including engineers with various lavels of experience, don't yet fully understand the implications of the open-eneded interface that language provides with LLM interaction.
AI is a helpline, so for a lot of people it cripples them. they cease to think. simple as that.
its something people need to be aware of that their brains work like this, because that awareness is the same thing that solves the problem. (people find it very confronting to learn they have been asking questions they knew the answers too already...)
Forwarding an AI response (or any canned response) sends the message that they don’t want to interact with you. That could be intentional.
> Don't post generated text or AI-edited text. HN is for conversation between humans.
When people start doing it in an entirely voluntary human-to-human situation, like having a conversation online about a topic of interest, I'm gobsmacked.
Sadly, the point is validating the human worth of the human who can't do any better than this.
But, assuming you're not stuck on that human's Self Esteem Care Team, it can be tricky to tell them that they need to seek validation elsewhere.
I can count on one hand the number of times I've seen LLMs one-shot responses that are fit to paste to Slack verbatim. It does happen though.
That time has now passed though.
Just have devs commit (or provide however you like) their ai coding sessions.
It then becomes obvious if someone has, or has not, reviewed the code, because then they discussed it.
Bonus points: on very complex tasks, you can more or less resume the conversation, via session files.
Same for schoolwork, research papers, and a lot more.
The conversation is not something to throw away: it is arguably as important as the code in the pr.
You can't just remember every conversation and ... something useful.
Indiscriminate memory buries you. What you need is "monumental" memories, which is also known as documentation. AI can be very effective in helping you manage that. That should be kept tight, just like code. You want to keep enough of a record of how you got to where you are that you are equipped to be effective in future tasks.
I would say that your suggestion is basically digital hoarding. Lets say there are 3 kinds of people, the person with OCD who keeps his house almost barren in the name of tidiness, the person who visits his possessions periodically with a critical eye, deciding what to keep and what to throw away, and the hoarder. When it comes time to solve some household problem, say hang a picture, which one of them will have what they need to do it (a hammer and a nail, maybe a little bit of wire) and be able to find it?
I think you might consider keeping the session history for a while, maybe a month or 3 months or 6 months, but have something that automatically deletes them after that.
Imagine your hoarder scenario except that the hoarder has unlimited storage space and a team of librarian/warehouse workers to go find and fetch the items needed for the wall hanging task.
I would totally lose it if someone come up to me with that shit. Zero chance I would put up with it.
This shit has progressed well past seriousness and into the absurd
I had a guy with more senior title say “I love the new api and cli. Is anyone working on the docs? Because I can take a crack at them” then goes to generate 12 markdown files, and shares them with me. All the cli commands and APIs are wrong. I was very confused. You have the api spec, and the cli. You couldn’t point your agent to them to validated the commands and the api? He replies “oh that’s a good idea. I just described them to the agent and thought you could review and fix mistakes”
It's not a bad time to take a break from it. Keep yourself from cognitive decline, then in 2 or 3 years when the industry has come to its senses you'll be able to impress in a sensible interview. Compared to the slop addicts that won't know how to tie their own shoes without asking Claude each time.
The effect is less pronounced in other industries but they're gonna catch up. Might take a decade for some of them , perhaps even 20 years for the rare ones like nursing but the trend is quite clear.
The system does NOT reward caring. That's the problem.
Oh, for the testing required by any customer of any software vendor any time from now to forever more, as a result of the possibility of this level of irresponsible outsourcing of brain.
Enshittification is a new word/concept, but it seems we've already reached the potential for enshittification squared.
It is clear though that virtually all degradation we have is in human capital and social sphere against the background of staggering technical capability.
I thought AI is an exponential?
I'm glad most of my colleagues are still doing things themselves. But honestly, if all you can do is tell me what Claude told you and you didn't even bother to read it yourself. Then I might as well just cut you off the team and go use Claude directly.
This is the worst person to get this from because they literally don’t understand it.
Everyone at my job uses LLMs to code, all day long. People still ask me "why does this bug happen" and I use Claude to find out and the answer is almost always straightforward, and when its not I raise a discussion with the team. The calculus isn't "can I prompt an LLM to find out," it's "do I have a solid enough mental model of the part of the codebase to prompt an LLM specifically enough, and also understand if the LLM is hallucinating or not."
> How can someone serious even think that’s ok.
I think this is a culture issue. I think every org is different, but we had a similar ramp up at our company. First we trialed cursor. Then some people were committing slop. Then we were like "llms are no excuse for slop." Then we got our skills and context good enough to not require close review. And then everything became LLM driven and now we have become "meat proxies" where everyone asks LLMs to fix things that are part of unfamiliar parts of the codebase and then asks someone else familiar to review the fix. We're tending towards a solution to this because I dont think its a good usage of resources but every org will come up with their own culture and solution to this.
It's hard to make something foolproof, because fools are so ingenious
LLMs have made G extremely cheap, but verification capacity and judgement has not scaled at anything like the same rate. Someone can generate a bunch of PRs or a report quickly, while transferring hours of verification work to everyone downstream.
We are on the wrong side of this ratio, because we never had the capacity to extinguish reviewer capacity the way we can today.
“Don’t be a meat proxy” needs to be a cultural norm for the larger problem of using cheap generation to externalise verification costs onto someone else.
I believe in proper breaks, more than enough downtime, work-life balance - but I’ve been hamstrung my entire career by people who seem to be so low-bandwidth that they often seem to be operating on autopilot - while making the same (or far more) as I’m making.
It’s the equivalent of choosing words from a conversation and sending you the dictionary entries.
I could do that myself. I really didn’t lack the ability to click a button.
Because no one is serious anymore.
I hate to say it, but anyone this lazy is absolutely asking to be replaced by AI.
I created a spreadsheet called “slop” and add the name to the list and redirect all queries to their engineering manager. I’m not wasting my time on that shit.
I think this is just the moment we are in, the amount of people using AI in their day to day work changed so quickly that we have a very wide spread range of competencies and experience and really nobody has had enough time to be an expert with AI. We don't have well defined best practices and it feels like the target is moving.
My guess is you were probably complaining about something similar to this before there was AI, this is just the new target for your angst.
Namely - I put near zero effort into this response, and deciding if it's correct is on you. -> I'm asking you to read it and figure out if it's correct because I couldn't be bothered to do so