Hah absolutely, after of a very intense project I’ve undertaken with Claude agent teams for a few months now, I have caught myself using opus expressions, since they come to mind now naturally because that’s most of the text I’ve read since then … I’m catching myself so far, but in a year or two maybe I will not, fable doesn’t do as much “mannerisms” and speaks plainly, so maybe by then we will all be running fable level models on hour local machines and just not notice it anymore?
Or maybe the frontier labs could run several "personality" variants of each model and route you to them at random. Sessions are already sticky because of cache so it seems like any given cluster could be assigned one out of several fine tunes.
Hah I think if a repo is “standard” enough that an agent can navigate it freely without any help or direction, maybe it’s not worth having altogether?
Even fable _regularly_ stumbles as big repos or custom configurations, even for projects that fable itself built with high dev quality standards and modern design direction.
It just can’t hold it all in its context and will be forced to do “software archeology” all the time to figure things out - yeah it will work _most_ of the time, but to truly be able to scale and have autonomous agents reliably work and mold your codebase you need a lot more structure - tests, lints, compilers, validators etc. Your “skills” or policy files are there so agents can resolve issues and heal things themselves without your explicit direction.
If I have several tabs, each holding an agent team, with each agent spawning subagents as it sees fit, all of that apparatus has to ground itself _somewhere_ and if you don’t make decisions yourself, it will make decisions for you, save them in its own skill files, but some of these you might not like.
Hmm that’s strange, in my experience it is the other way around - Claude is super diligent with infra and will _insist_ on double checking and trying everything for real before committing.
When I was doing this myself I would read the docs and just implement them - claud is going about doing real software archeology to figure if what is said is actually the truth or it’s stale/inaccurate/buggy.
I’ve become 10 times more diligent because it is a lot easier to do. It’s no longer Urgh it’s good enough let’s ship it, now it’s “sure put a leg on it to figure it out and double check it”.
Backups are _tested regularly_ now because LLMs make it cheap to do so.
The only problem is when new engineers who haven’t learned these things Pre-ai now don’t really get why it is needed in the first place and will often lead the agent astray.
I think to address this we need to change or improve our training routines in general for humans. I think a lot of companies nowadays just skip that and deploy a company wide skill/policy for the agents, but don’t transfer the underlying skills to the devs themselves.
> Backups are _tested regularly_ now because LLMs make it cheap to do so.
sighs heavily in 90's sysadmin
Testing backups is not just a question of whether or not the restore command works. Go back and read the Tao of Backup: http://www.taobackup.com/history.html . The application itself (in its current version, with its current features) needs to work with the backed-up data, and the only way to verify this is to attempt to actually work with the data.
If you don't trust your agent to ship to production without manually reviewing the output (in some way), you have no business trusting your agent managing your backups. The agent writing some tests doesn't mean that the tests adequately handle all of your actual scenarios, let alone that your system will adequately handle data that is missing since the last backup.
But isn’t “writing a linter for contracts” the thing we need and probably would solve?
Coding agents are great because they have compilers, linters, test cases etc to ground themselves in.
With tools like OKF I’m sure most knowledge work would be distilled to its core data - it’s AST if you will, and then allow models to guard against hallucinations.
Checking if a case law exists is a tool call, you can demand provenance, it’s all _buildable_.
Hallucinations are “solvable” this way, so the rest is just time and adoption…
I’m not convinced tbh, a lot of my colleagues messages are “x broken” or “I do y” … no context whatsoever, and then I have to coax their context for making these decisions.
Since people started using Claude now I get full context of everything… might be too much sure, but to be honest over-communicating seems better than under-communicating - sure it’s boring and tedious but that shifts the blockage to me.
Otherwise each of these coaxing sessions is something I have to keep in my head until resolved, which is a load in and off itself.
And even better - I can point Claude to that message and gives me a summary. You might say this is silly because we are paying the LLM tax, but knowing what to share, and then verifying against my situation are two different things.
Sometimes they have checked the wrong thing, sometimes they need guidance, sometimes I need to investigate before I can answer. A full LLM message with context is like “summary of their working context that I can resume on my end” so I don’t waste cycles asking or rechecking etc.
And it is still possible to be a learning experience for both of us since the initial message is the start, we can then talk to each other like humans, both much more in sync than before…
It is kinda ironic but I think it does push things in a better direction. Would I have loved it if the message was human, direct and to the point - obviously, but we are all busy, we got shit to do, and this is a useful “resume” mechanic
> Since people started using Claude now I get full context of everything…
And I think this is where one key argument on the page itself I think fails:
> The person on the other side has the same tools you do.
Yes, the person on the other side has Claude; but in many cases, their Claude doesn't have the same context as my Claude. Saying, "This thing happened in an area of the code I don't have any expertise in. Below is Claude's analysis, which I'm not qualified to assess" is I think a valid time to copy and paste LLM output. [1]
The other thing to say is that it can be a "LMGTFY" style response; i.e., "Why did you ask me that, when you could have gotten a perfectly good answer from ChatGPT?"
[1] In the particular case I'm thinking about, it was a bug in an ocaml component; I don't know ocaml or need to know it; handing Claude's analysis 'caveat emptor' to the maintianers who do know ocaml I think is more helpful than not.
The problem is when all we get is the output, we don't know if their context is even valid. If they don't demonstrate any human understanding of their message content, we can't really take that on faith. Especially once we've got them copy pasting blindly in the past.
Working in support most of my life has shown me the vast majority of customer context is invalid when written by humans.
People don't read error messages. They'll include exactly zero context in their tickets. They outright lie. You can't take anything they say with good faith until you have a relationship with that individual and discern their competence.
I think some of this is a large amount of HN posters are working on software teams that have communicate pretty commonly, hence have a shared context. Once you start getting outside of that shared context knowing what information to exchange gets vastly more difficult for humans.
> Below is Claude's analysis, which I'm not qualified to assess
I’m reading a bunch of game modding discords, and this exactly what happens, and what’s hated. Because Claude is also not qualified to assess it. Not by default. The modder’s Claude might be because of the extra context and steering, but the user’s produces usless slop, instead the actual log, and context what the user did is what’s helpful.
I’d say this (when you don’t understand it yourself) is pretty much the worst time to send AI output.
Quick edit: Yeah, I get it in situations like GP where the user is unhelpful, doesn’t want to learn, doesn’t care, and you also can’t tell them to fuck off. But just because the user gives Claude the context they should have given you in the first place, doesn’t make it good.
I agree that "The person on the other side has the same tools you do" might be technically true, I have myself encountered it often enough that they don't even use it at all.
People don't always want MY take on it, they just don't want to bother finding the solution on their own at all.
So in that case replying with a verbatim copy-paste of a Claude response has the same energy as a lmgtfy link.
My take is that the friction is the point, and I'd much rather have a colleague communicate poorly to me than have them give the LLM response. The reason being I view my relationships at work as investments, and working together (communicating) helps build that investment through learning and feedback.
If someone says "x broken" and that's a bad response it becomes a learning opportunitity, something I can invest in. When I communicate poorly, I hope my colleagues help me understand why and give me pointers to improve.
Perhaps I'm being overly optimistic, but I've had the best growth with co-workers when actually engaging at a meta level and trying to help (and be helped) communicate better.
Using AI shifts that burden and entirely demoralizes any investment motivation I have. It's a view that communication is transactional rather than collaborative (we're all on the same team).
I want my team to work better, and be better, and if people offload their comms to an external system to generate I've lost that opportunity for growth.
The difference with AI is that the agent becomes the first level of triage. It (many times) asks the easy initial questions (did you turn it off and on again?) so I don't have to. Doesn't always work like this, but often enough.
If everyone thinks their question will get closed, it might as well be reality because no one is going to ask and they are going to go elsewhere. Which is exactly what happened.
The irony is that this is the exact opposite of the "let me google that for you" meme of the previous decade. If you have a question that could've been answered by AI, maybe you should've used AI in the first place.
I have the totally opposite sentiment from the previous LMGTFY meme. It was easier before to ask humans rather than do some research using Google, that's not true anymore with LLMs.
Now if I ask a question to humans I'm looking for their personal opinions drawn from their personal experience, be it ticket size for a feature on a legacy codebase, a design decision or a product strategy decision.
I agree that if someone asks you something, you should at least write "I asked AI and it said..." rather than pretending the AI response was your own. If you ask me something and my first response is to ask AI, it's probably because I only know enough on the subject to ask AI, and don't know much more beyond that, which would be the real answer.
I agree. Poor communication is one of the most expensive and frustrating things about life and society and business and development. LLMs are helping people better structure their thoughts and that's fantastic. Sure, it can feel impersonal, but once you get past that, I think the benefits far outweigh the costs.
The exceptions are spam, advertising, and propaganda. Because the barrier to writing now is effectively zero, there is no cost to writing unlimited tripe. So LLMs work well if the parties know each other is real. However I think it poses a major threat to the existing internet as it stands.
Hah, I guess we are not far away from the “Avogadro Corp” future - and that book was written just _before_ LLM, chatgpt and the attention is all you need paper :) That author had quite the insight
> Points to the future where my bots talk to your bots. I think that’s healthy.
I can't imagine why. You have become a robot. You've automated yourself out of existence. And due to your consistently mindless replies, everyone now knows there's no point in talking to you ever again.
1. I've experienced both (ai paste slop & ai providing helpful context that would otherwise have been absent) but the former has been a much bigger problem than the purported benefits of the latter
2. before the advent of LLMs, I had fairly strong contrarian opinions on the latter (clueless users asking questions with no detail) that I still hold to this day. A good proxy for my opinions on this is that I think https://nohello.net/ is a crutch for undersocialised (or overworked/near-burnout OR undiagnosed autistic) engineers & should not be normalised. The origin of this sentiment is users entering a quiet IRC room & saying "hello", then leaving & not being around when someone eventually replies: in that context nohello is pragmatic, but taking it beyond that to modern high-frequency realtime chat is an antisocial overstep. Leading with the entire context of your query may have some practical benefits but they're honestly overstated in exchange for killing natural social tendencies towards human dialogue. Similarly, a non-technical user giving you zero info about a problem is an opportunity for conversation & mutual education. It's relationship-building & helps build a foundation for further interactions with this person. It was never a problem in the first place; "solving" it is just another milestone on the road toward de-socialising humanity.
Well I mean… you didn’t come into this comment section saying “hello” and then waiting for someone to say hello back before commenting, right? And you don’t do that in emails either, right?
Because that would be super out of place, and you understand that intuitively.
That’s all nohello is about. It’s not trying to say “don’t be nice and say hi”, it’s saying “batch your hello and your question in the same message please”. Asynchronous communication mediums need an asynchronous communication style, it’s not that hard.
I've generally seen it used in Slack, which can be sync/async depending on context, but most commonly sync. I've never seen it used in online forums or e.g. email so not really sure what you're getting at with your comment.
Either way it was intended as an analogy - being annoyed that a nontechnical user isn't given you everything up front (because they don't know how) is essentially just being annoyed with having to talk to somebody. The solution to that isn't something the nontechnical user should be burdened with, imo, & doing so doesn't tackle the crux if the original issue in any meaningful way.
> Slack, which can be sync/async depending on context
Right, the nohello people are saying “don’t treat initial communication on Slack as sync.” At least not until after the conversation has started. Since you don’t know if I’m available to reply yet, just batch your question and the pleasantries together. It’s a super easy idea to grasp and it really ought to be universal. Just batch the question and the pleasantries together.
> I've never seen it used in online forums or e.g. email so not really sure what you're getting at with your comment.
Right because forums and email are more generally understood to be async. Slack has a lower latency for a message to be delivered, which can enable synchronous communication (for people who can type really fast at least) but it doesn’t mean you should just assume someone is going to be able to reply at typical synchronous communication times: they may be busy, they may be talking to someone else, they may have stepped away for a second. So until the conversation is “underway”, treat it the same as email and forums. Batch the pleasantries and the question together.
> being annoyed that a nontechnical user isn't given you everything up front (because they don't know how) is essentially just being annoyed with having to talk to somebody
It has zero. Zilch. Nada. Nothing. To do with technical anything. It would be exactly as important for, I dunno, sales reps to use this communication style between themselves. Or between a clothing tailor and a client. Or between, I dunno, farmers or something. (For scenarios you could imagine them using Slack at least.) In every case, if you have something you need to ask, you can just ask it in the first message. You don’t have to wait for someone in the channel to acknowledge your presence via pleasantries first.
It’s not about being “annoyed with having to talk to someone”, it’s about not wasting people’s time. It matters in every place where people’s time is valuable. So, everywhere.
(To extend an olive branch I’ll posit some places where the hello/pleasantries style is totally acceptable… group chats among friends or family, purely social channels, anywhere where the purpose of the communication is the pleasantness of the communication itself, ie. leisure.)
> The origin of this sentiment is users entering a quiet IRC room
For me, the origin of this sentiment is the culture for certain departments at $LARGE_CORP I used to work for. People would ping me on the internal instant messaging tool to ask for help with things. Sometimes they'd take the nohello approach and just fold everything together in a single message: "Hello, hope you are having a nice day. I'm trying to do X, and I see Y instead of Z. What am I doing wrong?"
And sometimes, they... don't.
"Hello." <wait>
"How are you today?" <wait>
A conversation that could have been done in 30 seconds drags on for many minutes, and the message trickle feeds just fast enough that I can't do anything of any complexity in between the IM app's alert lighting up again. Extracting what they actually want from them feels like pulling teeth. It's a complete waste of their time and utterly disrepectful of mine.
They are there to ask a question. I know they are there to ask a question. They know they are there to ask a question. I'm happy to answer the question, when they finally get around to asking it. Just ask the bloody question, ffs.
You are right - and this has been the case for me before, but unfortunately human resource constraints are not something most orgs optimise for, including my current one.
I would love to spend more time and talk to the people with their issues so we can come to a common understanding… but neither have the capacity for this, and now things are getting worse I think.
Maybe someday agents are good enough that we just discuss direction and implementation details, and every “developer” becomes more of an executive in a company, free to chat about more than just work, focusing on building human relationships and leaving “the tech stuff” to their agent teams…
And old geezers like us who actually know how to write ourself out of a for loop would be the premium - wizards holding forgotten knowledge, incomprehensible to mere uninitiated mortals. I wonder what distinctive dress and hair styles will we develop :-D
If you're absolving the other person from any understanding of the issue and the alternative is "stuff broken" then yes what you say makes sense. But that's an incredibly low bar and one would have to say - that's not a colleague, that's an end user type relationship.
> a lot of my colleagues messages are “x broken” or “I do y”
These people had below than average communication skills (or at least they were putting below than average effort communicating with you), so LLMs improved their output.
However, the society kinda expects people can and Should improve themselves over time. The issue is that LLMs improved their output, but not necessarily them.
And did the LLM really improve their output? They gave similar output to the LLM, so if it doesn't have a harness with tools connected to fill in the blank, it might be hallucinating details. Might not even count as a hallucination as it tries to fill in the blank with the closest looking relevant information (previous chat, maybe it has access to teams/emails and scans that for anything similar, and so on).
I've sees some chat logs where I completely empathized with the AI trying to make sense of the information it was being drip fed. Always incomplete, often incorrect.
> I’m not convinced tbh, a lot of my colleagues messages are “x broken” or “I do y” … no context whatsoever, and then I have to coax their context for making these decisions.
> Since people started using Claude now I get full context of everything… might be too much sure, but to be honest over-communicating seems better than under-communicating - sure it’s boring and tedious but that shifts the blockage to me.
Yes, but the kind of engineer that used to say "X is broken" will send you some slop with missing context and Claude will try to apply general solutions from its training data rather than from the specific context.
As somebody on the Ops side that usually helps devs fixing things here and there, in those cases I'd rather them say "X is broken" and then I ask the question to Claude with all the context I have.
Probably the best solution currently is a shared skill with at least some curated context.
> sure it’s boring and tedious but that shifts the blockage to me
I think that's the key. Generally people would rather not receive slop because the receiver has no idea how much thought the sender invested in the communication.
In your example, the more context the better. I think there's an interesting nuance there.
> the receiver has no idea how much thought the sender invested in the communication.
That's an interesting thought. The message can have metadata that says the sender's spent 1/5/10/60/120 minutes or some other metric of how much thinking over the communication happened. Just need a way to monitor people's thoughts and digitize them to make that happen.
Don’t coworking spaces solve some of the issues? I am one of these people that hates working from home and loves going to the office - the chat with colleagues, the exercise of bicycling to the office and back, the context switching between work and not work - otherwise I loose myself in the work. All of that is awesome for me.
When I was playing around with remote work, (before the pandemic and before it was cool) - I figured most of this can be offset by just going to a coworking place - and even better - friendships you form from there seem stronger, because people are less anchored to the company and more to the “place” - more open to friendships. Plus you can pick and switch your “office” as desired for your mode of travel and preferences (or even switch cities, travel around)
Everyone is always a squatter on someone else’s land, for millennia now.
The question is how do you deal with it. Israel “dealing” with its “Palestinian problem” by taking a book out of nazis and Ottoman Empire is both ironic and incredibly sad.
Also kinda inevitable if you think about it just from a game theory point of view. Every act of violence, every war, won or lost, increases support for a radical party. Every act of terror Hamas / Hezbollah promote strengthens Netanyahu’s position, every act of terror Israeli government does strengthens Hamas / Hezbollah.
It’s just game theory - either respective positions are not to “solve the actual problem” it’s to perpetuate their own existence. If the terror stops, all three lose power, so which leader would ever do it?
There has to be either an outside enemy to drive people together, or an outside force of peace doing the same (EUs role in the Northern Ireland’s case). And this becomes even harder if you have external powers that are _driving_ the division (Iran for example in be middle eastern case).
It’s just humans being greedy and thinking of local political problems and not the bigger picture, a “premature optimisation” if your will…
It's incredibly cynical to demonize Jews for living in our historic homeland and turning desert sand into thriving cities[0]. It's what we prayed for daily for centuries after being thrown out of Europe. Now it's here, now we're living it, and haters call us by our persecutors' name for doing it.
Tel-aviv is honestly incredibly impressive, I’ve been there and the place is amazing - that it was just a sand dune and without the Arab oil wealth still managed to make this land into an amazing place to live…
Don’t think anyone is faulting people for that - it’s the other side of the coin that the whole world is kinda not really liking.
And it’s not like there aren’t examples of successful integrations - even with internal tensions some places do manage to just make people live together, it’s hard yes but not impossible… just have to have the will and smarts for it, wonder what the current Israeli government is lacking…
How should they have dealt with it? The responses have been pretty reasonable. Even 2023 was a reasonable response. Only as the war dragged on did it go to far in my opinion.
Honestly, I’ve stopped caring about code readability for a few months now. I want the code readable _to the agent_ not so much to me.
I don’t trust it with code anyway - every feature needs comprehensive test, and then a live deploy on a real working test system before it is approved - I mostly measure success with - after deployment is it doing what it’s supposed to be doing.
It’s like “helping another team managing their work stream” experience rather than coding yourself.
Funny enough models seem to have personalities and the dis on each other - when I had an opus orchestrator dispatching fable workers, they would comment on how “unreliable” it was and it had “evidence to prove it” and fable thinks opus is too rigid and needs more hand holding… it really starts to feel like managing team egos and verifying work.
And I code scan mostly to just spot check if it’s not doing anything super stupid. But my goal is to make sure anything shipped is easy to change and fix, and every mistake has a test behind it so it doesn’t happen again.
I ship more problems, but they get discovered and fixed quicker. Before they reach prod of course.
And from time to time you do reorganisation and refactoring passes where I brainstorm how things could have been better with the help of evidence- chat sessions, tests, bugs etc.
It feels less like rigorous engineering and more like gentle gardening. Or like “project management” not “coding”.
Honestly given my age now I’m fine with that. Have enough “hard” projects under my belt (ORMs, sql parsers, etc) that I don’t feel I need to prove anything to anybody, but I don’t think that’s even relevant- the velocity change is … I guess around 5-10x for me - with provable metrics, so I try not to lent the good old days but figure out how I can now live in this brave new world and be happy with my work.
Hmm just shared this with someone who works in this space in India, they got angry.
> It doesn’t work for 3rd world contaminants.
> White people have been making these for years. Doesn’t work
> We need hard TDS removal also
And the ceramic ones apparently are also bad:
> we had one of these for many years while growing up. the ceramic candles got fungus on them
Viruses are the bigger threat I think and just a filter like this will not stop them HepC and the like would be easily transmittable. Not really “safe”, maybe at best “better than nothing”
She is leading some orgs in rural India, targeting empowering and helping women and has seen countless examples of Europeans / Americans coming over with noble ideas, making a mess, leaving, and sometimes things getting worse not better in response.
Took her a lot of empathy and understanding on what the situation is like on the ground really, second / third order effects, etc - to get to solutions that _actually_ help people there.
For example how do you do fair lending to people (rural Indian women) who have no bank account, no credit history, almost no financial history at all, and they want to star an enterprise? How do you “bootstrap financial trust?” And with proper and inventive organising she managed to lend to women and proved with data that they are 5x better borrowers than the equivalent male population, but were starved of accessible credit cause banks would not lend to them.
reply