> The success of Claude Code and Cursor at the higher end of the market shows that even the people pickiest about their software (developers) will use your software regardless of how good the code is.
Seems wrong. Devs will whine, moan and nitpick about even free software but they can understand failure modes, navigate around bugs and file issues on GitHub. The quality bar is 10-100x amongst non-techno-savvy folks and enterprise users that are paying for your software. They’re far more “picky”.
WOW. That does really drive home the perspective. I was an adolescent during those years and it did seem quick then, but that's an insane pace in retrospect.
Amazon is perhaps a counter-example to your point, though, to be fair. It seems to me they did a lot of spaghetti throwing while making accounting losses for a good number of years. Granted, they did it on OpenAI's dining budget.
People are much more willing to give the benefit of the doubt on things like that when the flagbearers of your industry aren't running around sucking all of the oxygen out of the system and telling people things are "solved": that your product will obsolete them in the next 6-12 months.
We get it. They say that stuff to raise money, make sales and keep the party going. But don't expect too much sympathy when the strategy falters a bit.
I mean, tools change, but I'd be happy to hear if any tool can create that by just saying create "Claude Code Unpack" with nice graphics. or some other single prompt. It likely was an iterative process and it would be lovely if more people started sharing that, because the process itself is also very interesting.
I've created some chinese characters learning website and I took me typing 1/3 of LoTR to get there[1]. I would have typed like 1% of that writing code directly. It is a different process, but it still needs some direction.
I think it is accurate. Where are the autonomous AI who beat the creator to the punch? When we write "Hello, World!" in C and compile it with `gcc`, do we give credit to every contributor to GNU? AI is a tool that thus far only humans are capable of using with the unique inspiration. Will this change in the future? Certainly. But is it the case now? I think my questions imply some reasonable objections.
It’s this everywhere: the constant fear of not raising children perfectly in every aspect puts downward pressure on a family’s desire (and perceived ability) to have more children.
I don't think it is about raising children perfectly. I think people knowadays are more educated about negative psychological impact to children and therefore there is a change in parenting style which takes more effort. More quality time, no corporal punishment, more understanding and kindness and so on. The effort to raise children therefore has increased (fear based parentingis rather low effort). I think it is for the good. Those who are able to make that commitment raise the kids. (I dont say that negative styles don't exist, but society in general has improved on parenting styles)
I work mostly on the tech side of things but my corporate limitation has always been writing up documentation, communicating/translating to stakeholders, and recalling everything relevant when writing PR descriptions. AI has been a breath of fresh air. I actually communicate more information efficiently than I would have ever put the effort into before. I still maintain my own writing for more casual things like social media (HN included) and low stakes Slack conversations but AI for getting across ideas and then proofreading it is great.
"I actually communicate more information efficiently than I would have ever put the effort into before"
- this is subjective and evidence seems to point to the opposite in my view. In reality most people who think they communicate better with AI don't actually read what the AI has written for them and just puke it out on the world, expecting their readers to do the work.
The Ai almost always writes boring, repetitive garbage and very, very often includes redundant information. But saying it creates more efficiant communication is a great excuse for being sloppy and lazy.
I have had the same experience, personally. i.e. asking Claude to simplify things for c-suite has gotten (1) extremely positive feedback from c-suite and (2) actually relevant conversation about decisions. It's certainly not a one-shot but iteration with Claude is so fast that it takes just a few hours vs plotting weeks about how to clarify technical decisions. But Intend to work in a "try it this way" sort of iteration where I need to rewrite things and see what they look like. But using Claude/ChatGPT for options about whether things make sense is very helpful (for me). The speed of iteration is great.
Which one is it? Subjective or evidence based? I'm sharing what I know is true for my experience as well as the fact that I proofread what I send with AI and am aware of how terse I usually am.
I was asked to write user stories about a complex topic where I’m the SME at work. I spent two hours info dumping everything I knew about the project, everything the AI wouldn’t have any context for, using Cursor to add related projects to the workspace, tagging specific files where we’d implemented similar things with our styles, noted all the quirks of the system and how it works and where to find relevant information. I spent a lot of time on it, and then asked it to reach out using cli to grab relevant information from our infra, and write stories about how we’d accomplish everything I intend to get done. I then spent another few hours reviewing the 45 or so stories that conversation generated. It was similar to how I’d talk to a new contractor I’m onboarding to work on the work.
I have a deep knowledge of the information, have done the process we’re doing on two previous projects, but organizing all the stories would have been an absolute nightmare. I still spent half a day on this, I’d guess the fatigue from the boring parts would have made this take a week or maybe two, just because I was doing the parts I enjoy (knowing things and describing them) and I was able to offload the parts I’m not great at (using a lot of boilerplate language to organize the info I knew into scrum stories). Then I had a meeting, reviewed the stories with my coworkers, we had a discussion, deleted two or three of them that we determined weren’t necessary, and fixed up one or two where I’d provided insufficient information about some context surrounding coloring of a page.
It burned through a ton of Opus 4.6 tokens, looked through a ton of code (mostly that I’d written, pre-LLM), but has been amazing for helping me move into a lead position where grooming stories and being organized has always been my weakest point.
Also, when I wrote a postmortem for a deploy that had some issues, I wrote it all by hand. You have to know when the tools help and when they will hinder.
I thought it's quite good. Of course, I'm not taking 100% of output, but it takes care of my grammar blindspots (damn you commas and a/an/the articles!).
Can you please share what and how gets degraded? Sometimes I don't like a phrase it selects, but it's not common
> it takes care of my grammar blindspots (damn you commas and a/an/the articles!)
There are plenty of pre-LLM tools that can fix grammar issues.
> Can you please share what and how gets degraded?
I'm not the person you asked, but IMO LLMs suck the style and voice out of the written word. It is the verbal equivalent of photos that show you an average of what people look like, see for example:
As definitionally average the results are not bad but they are also entirely unremarkable, bland, milquetoast. Whether or not this result is a degradation will vary, of course, as some people write a lot worse than bland.
Well, for one example, it inhibits your desire to improve against those very blind spots. In exchange for that your audience gets 3-4x length normalized bullshit to read instead.
AI can take a rough draft, clean it up and shorten it as much as you want. The suggestions very often expose ambiguities in the original text. If you think the LLM got it wrong, it’s nearly often the LLM overreading some feature of the original that you failed to catch, which is precisely what you’d want out of your proofreader.
Yes, LLMs reduce the individual charm of prose, but the critique itself carries a romantic notion that we all loved the idiosyncratic failures of convention and meaning which went into highly identifiable personal styles, and which often go missing from LLM-edited work.
> Well, for one example, it inhibits your desire to improve against those very blind spots.
I'd contend this is not true. Even professional authors go to an editor who identifies things that need to be fixed. As the author of the text and knowing what it should be, it can be difficult to read what you wrote to find those mistakes.
> In exchange for that your audience gets 3-4x length normalized bullshit to read instead.
This is not at all what is implied by having an AI act as an editor. Identifying misplaced commas, incorrect subject verb agreement (e.g. counts), and incomplete ideas left in as sentence fragments.
You appear to be implying that the author is giving agency to create the content to the AI rather than using it as a tool to act as a super-charged grammerly.
> Even professional authors go to an editor who identifies things that need to be fixed.
Yes, and these people are good at it. What’s your point?
If you need grammar checking, there are thousands of apps including word processors, web browsers and even most mobile devices that will check your inputs for grammar and spelling mistakes as you type. All of that without burning down the rainforests or neutering your thesis.
I believe you are confusing what an editor does and proofreading.
In the time before LLMs, for some of my occasional blog posts I'd first post it to whatever messaging system my colleges used and ask them to read over it. Identifying that "this word is confusing in this context" or "you're using jargon here that I'm unfamiliar with" is helpful. There's also stylistic items of "this sentence goes on for far too many words and thoughts without making a single punctuation mark indicating where it is complete or delineating two or more different ideas leading the reader to have to keep back tracking the thought to try to keep it all in their mind which can be confusing and makes it more difficult to read."
Proofreading tools pick up some typos and punctation errors in that previous bit. https://imgur.com/a/oqqoEGV None of them called out its structure.
The overly long example sentence introduces unintended humor or self-parody, which may dilute the seriousness of the point.
Now, one could argue that taking its advice for the structure and that I have incompletely formulated some arguments would change the tone of my writing. However, any changes that I make are changes that I intend to make and are not the result of the LLM rewriting my words.
In many kinds of writing, perhaps most, communicating your state of mind to the reader is a primary goal. Even a smart LLM fundamentally degrades this, because to whatever degree that it has a mind it isn't shaped like yours or mine. I've had a number of experiences this year where I get to the end of a grammatical, well-structured technical document, only to find that it was completely useless because it recited a bunch of facts and analyses but failed to convey what the author was thinking as they wrote it.
(Of course, that may well be exactly what you're looking for if you're writing an audit report or something.)
This sounds like an ESL issue. LLMs are good at proof reading ESL-written English text. They are not as good at proof reading experience English writers.
AI for editing is good and have many useful cases. The part where it fails is that the tone/style of the writing gets overtaken and reads like all other AI edited writing. But the quality of the edit is good, its just not in your style. When everyone sounds the same then there is no uniqueness. But using it edit legal letters, software documentation etc are very good use cases, using it to explain your ideas in a blog not so much.
Depends on how you use it. If you say "reword this to sound <whatever your goal is>" then it does suck. But if you say "This is what I wrote. My intention is so-and-so. The audience is <audience>... Please mention and add suggestions for how to fix typos, poor wording, unclear expression, etc.
Then you get back what it thinks is wrong and you're in charge of editing in its suggestions. If you let it edit for you you're more likely to just create slop.
---
Here's an example. My actual text is:
> I want to make it clear that I'm not hunting for things to be angry at, these are issues I've encountered in actual codebases.
If go through the route of prompting it to re-write, it changes it to:
> I’m not looking for things to be angry about—these are issues I’ve encountered in real codebases.
The em dash is a clear give away that it's AI, but it's also soulless.
If I ask it to tell me what's possibly wrong about it I get that there's a comma splice (never knew the term, I'm not a native speaker) and "about" is better than "at". So I do a minor change:
> I want to make it clear that I'm not hunting for things to be angry about. These are issues I've encountered in actual codebases.
It’s kinda useful to me for the following three reasons:
- spelling
- grammar or weird grammar as English is not my native language
- read proofing and finding things that do not make sense in terms of sentence structure
I do not use it for ideas, discussing the writing, or anything else because that beats the purpose of writing it myself (creative writing).
Only if you don't understand how to control AI. If you understands how it works and have the skills to ride it like a wild horse, you can make yourself a 10x developer. Its maybe a bit of an insult, but you seriously have to change that mindset. AI is not going to be worse tomorrow. It will get better and it will dramatically change our life as developers. Code will no longer be a prominent thing we are working on in the near future.
> They lose a big customer for their cloud services. Even worse considering that now, using the AI they helped fund, everyone can compete with their sub-par products. GitHub is a good candidate for disruption, and that’d be just the start.
Look, I'm a Microsoft hater like the rest of us, but calling Microsoft's products sub-par discredits the author a good bit. I invite anyone who thinks this to try and compete with them. Go after something like Word, for example. Then prepare to be awed by what some of the most brilliant programming minds ever can produce after grinding for four decades.
If I saw a helicopter crashed into a tree, I don't have to be a helicopter pilot to know it's not an ideal state of a helicopter and something/some people failed.
When I'm using MS Word and it takes 20 seconds to cold launch on a machine that's magnitudes faster than any computers 25 years ago where it launched near instantly, I can tell something is going wrong. When all of their software is harassing me to use AI in ways I don't want to use it, I can tell something is going wrong.
Microsoft's products do not occasionally fail, they're constantly going out of their way to block users from doing basic tasks through ads and dark patterns. It makes some KPI go up so some asshat product manager can get a promotion, and they never lose users because 99% of their users are hostages.
You can have an opinion about a tool as a user, without ever having ability to create such a tool yourself, that's literally what every tech and auto reviewer does.
I'm sure Word is full of arcane backwards compatible tricks that 20% of users use, but I find it hard to differentiate the Pareto 80% of the product from Google Docs or any other competitor (LibreOffice?) Adding rich text, tables, headings and colors is pretty much a solved problem for all of these softwares. Adding images or handling more complex layouts sucks everywhere, it's not like that Word has a great user experience and the other don't. All of them are bad.
IMHO, if we had any of the competitors being the de-facto standard for word processing, the vast majority of users wouldn't feel the difference. Power users would for sure, but I'm not sure they're many or they use existential features. If Word didn't have a near monopoly in office settings due to aggressive marketing, OS presence and a proprietary file format that constantly changes and never renders well outside of Microsoft products, it could disappear without anyone (save Microsoft) losing much.
Yes. That 80% you find useful is served fine by Google Docs, but there’s a good reason the enterprise overwhelmingly goes for Word, and it lives deep in that 20% and a lot of the time has zero overlap with others.
Microsoft's AI, on the other hand, is underwhelming at the moment and might well go the way of Windows Phone. Plus enough people hate the copilot icons everywhere that Microsoft is hinting at dialing down a bit.
MS Office should last a while if they stop calling it "Copilot 365 Office" or whatever it was.
I think Github represents 'par'. Plenty of stuff worse and plenty of stuff better. Overall it's what most people expect a coding social media site to be because it set those expectations. Those of us who are only looking for code management (including issues/PRs/etc) are easily satisfied elsewhere.
There are some frustrating parts, but subpar is an odd way to describe GitHub to me. I’m pretty happy with what they’re doing, and find the UX super helpful. I do agree Actions needs a debug mode but otherwise I get a ton of value out of the service for $20/month?
> LLMs are the new reality, they are not going to go away
That's the conventional wisdom, but it isn't a given. A lot of financial wizardry is taking place to prop up the best of these things, and even their most ardent proponents are starting to recognize their futility once a certain complexity level is reached. The open weight models are the stalking horse that gives this proposition the most legs, but it's not given that Anthropic and OpenAI exist as anything more than shells of their current selves in 5 years.
But LLMs themselves are literally not going away, I think that's the point. Once a model is trained and let out into the open for free download, it's there, and can be used by anyone - and it's only going to get cheaper and easier.
Yeah like Kimi is good enough, if there was some kind of LLM fire and all the closed source models suddenly burnt down and could never be remade, Kimi 2.5 is already good enough forever.
Good enough is probably redundant, it's amazing compared to last year's models
Makes some sense to me, as the prisoner's dilemma dictates at least some fraction will try to kill you. So you've got to go first.
Reminds me of the Dan Carlin take on aircraft carriers in World War II: if you in a carrier spotted an opposing carrier and didn't send everything you had before it spotted you, you were dead. The only move was to go all in every time.
Surely that logic applies if you're at war with the other side? If you start attacking carriers of random superpowers, you only invite further destruction...
Yes, but in this case the aircraft carriers represent the entire civilization and there is no repeat player such as nations in times of peace. So if you spot an alien civilization with even a 1% probability of trying to kill you, you don't really have the opportunity to "wait and see". And there's no meta higher-level arrangement that will protect you.
Seems wrong. Devs will whine, moan and nitpick about even free software but they can understand failure modes, navigate around bugs and file issues on GitHub. The quality bar is 10-100x amongst non-techno-savvy folks and enterprise users that are paying for your software. They’re far more “picky”.
reply