[hemloc_io]

The most frustrating thing to me about this most recent rash of biz guy doubting the future of AI articles is the required mention that AI, specifically an LLM based approach to AGI, is important even if the numbers don't make sense today.

Why is that the case? There's plenty of people in the field who have made convincing arguments that it's a dead end and fundamentally we'll need to do something else to achieve AGI.

Where's the business value? Right now it doesn't really exist, adoption is low to nonexistent outside of programming and even in programming it's inconclusive as to how much better/worse it makes programmers.

I'm not a hater, it could be true, but it seems to be gospel and I'm not sure why.

Mapping to 2001 feels silly to me, when we've had other bubbles in the past that led to nothing of real substance.

LLMs are cool, but if they can't be relied on to do real work maybe they're not change the world cool? More like 30-40B market cool.

EDIT: Just to be clear here. I'm mostly talking about "agents"

It's nice to have something that can function as a good Google replacement especially since regular websites have gotten SEOified over the years. Even better if we have internal Search/Chat or whatever.

I use Glean at work and it's great.

There's some value in summarizing/brainstorming too etc. My point isn't that LLMs et al aren't useful.

The existing value though doesn't justify the multi-trillion dollar buildout plans. What does is the attempt to replace all white collar labor with agents.

That's the world changing part, not running a pretty successful biz, with a useful product. That's the part where I haven't seen meaningful adoption.

This is currently pitched as something that will have nonzero chance of destroying all human life, we can't settle for "Eh it's a bit better than Google and it makes our programmers like 10% more efficient at writing code."

[aurareturn]

  Where's the business value? Right now it doesn't really exist, adoption is low to nonexistent outside of programming and even in programming it's inconclusive as to how much better/worse it makes programmers.

I have a friend who works at PwC doing M&A. This friend told me she can't work without ChatGPT anymore. PwC has an internal AI chat implementation.

Where does this notion that LLMs have no value outside of programming come from? ChatGPT released data showing that programming is just a tiny fraction of queries people do.

[hshdhdhj4444]

> This friend told me she can't work without ChatGPT anymore.

Is she more productive though?

People who smoke cigarettes will be unable to work without their regular smoke breaks. Doesn’t mean smoking cigarettes is good for working.

Personally I am an AI booster and I think even LLMs can take us much farther. But people on both sides need to stop accepting claims uncritically.

[fragmede]

> Doesn’t mean smoking cigarettes is good for working.

Fun fact; smoking likely is! There have been numerous studies into nicotine as a nootropic, eg https://pubmed.ncbi.nlm.nih.gov/1579636/#:~:text=Abstract,sh... which have found that nicotine improves attention and memory.

Shame about the lung cancer though.

[iberator]

Nicotine does not cause cancer. Smoke do

[phatskat]

Yes however nicotine can speed up the growth of existing cancers.

[legucy]

Cigarettes were/are a pretty lucrative business. It doesn’t matter if it’s better or worse, if it’s as addictive as tobacco, the investors will make back their money.

[KoolKat23]

Productive how and for who?

My own use case (financial analysis and data capture by the models). It takes away the grunt work, I can focus on the more pleasant aspects of the job, it also means I can produce better quality reports as I have additional time to look more closely. It also points out things I could have potentially missed.

Free time and boredom spurs creativity, some folks forget this.

I also have more free time, for myself, you're not going to see that on a corporate productivity chart.

Not everything in life is about making more money for some already wealthy shareholders, a point I feel sometimes lost in these discussions, I think some folks need some self-reflection on this point, their jobs don't actually change the world and thinking of the shareholders only gets you so far. (Not pointed at you, just speaking generally).

[johnnyanmac]

>Productive how and for who?

For me, quality is the biggest metric, not money. But time does play into the metric of quality.

The sad reality is that many use it as a shortcut to output slop. Which may be "productive" in a job where that busywork isn't critical for anyone but your paycheck. But those kinds of corners being cut seems anathema to proper engineering or any other mission critical duties.

>their jobs don't actually change the world and thinking of the shareholders only gets you so far.

I'm worried of seeing more cases like a lawyer submitting cases to a judge that never existed. There's ethical concerns about the casual chat apps, but I can leave that to others.

[KoolKat23]

I think this is not really the case, people see through that type of LLM use immediately (busywork). This is demonstrated in the fact that top-down implementations aren't working despite use amongst employees thriving.

People doing their jobs know how to use it effectively. Just because corporates aren't capturing that value for themselves doesn't mean it's low quality. It's being used in a way that is perhaps reflected as an improvement in the actual employees standing, and could be bridging existing outdated work processes. Often an employee is powerless to change these processes and KPI's are notoriously narrow in scope.

Hallucinations happen less frequently these days, and people are aware of the pitfalls so account for this. Literally in my own example above it means I have more time to actually check my own work (and it's work) and it also points out factors I might have missed as a human (this has absolutely happened multiple times already).

[sotix]

> Doesn’t mean smoking cigarettes is good for working.

Au contraire. Acute nicotine improves cognitive deficits in young adults with attention-deficit/hyperactivity disorder: https://www.sciencedirect.com/science/article/abs/pii/S00913...

> Non-smoking young adults with ADHD-C showed improvements in cognitive performance following nicotine administration in several domains that are central to ADHD. The results from this study support the hypothesis that cholinergic system activity may be important in the cognitive deficits of ADHD and may be a useful therapeutic target.

[johnnyanmac]

So the best interpretation is that it's like Adderal. Something to be carefully prescribed to with doctor-sanctioned doses. Not something you buy off the counter and smoke a pack a day of.

[aurareturn]

No she’s less productive. She just use it because she wants to do less work, be less likely to get promoted, and have to stay in the office longer to finish her work.

/s

What kind of question is that? Seriously. Are some people here so naive to think that tens of millions out there don’t know when something they choose to use repeatedly multiple times a day every day is making their life harder? Like ChatGPT is some kind of addiction similar to drugs? Is it so hard to believe that ChatGPT is actually productive?

[freehorse]

It is the kind of question that takes into account that people thinking that they are more productive does not imply that they actually are. This happens in a wide range of contexts, from AI to drugs.

[1123581321]

It isn’t a question asked by people generally suspicious of productivity claims. It’s only asked by LLM skeptics, about LLMs.

[happymellon]

It absolutely is a question people ask when suspicious of productivity claims.

Lots of things claim to make people more productive. Lots of things make people believe they are more productive. Lots of things fail to provide evidence of increasing productivity.

This "just believe me" mentality normally comes from scams.

[johnnyanmac]

>It isn’t a question asked by people generally suspicious of productivity claims.

Why not? If you ever got an AI generated email or had to code-review anything vibecoded, you're going to be suspicious on who's "more productive". I've read reports and studies and it feels like the "more productive" people tend to be pushing more work onto people below or beside them to fix the generated mess.

I do believe there are productive ways to use this tech, but it does not seem like many people these days has the discipline to establish a proper workflow.

[rimunroe]

That doesn’t seem to me like a good reason to dismiss the question, and especially not that strongly/aggressively. We’re supposed to assume good intentions on this site. I can think of any number of reasons one might feel more productive but in the end not be going much faster. It would be nice to know more about the subject of the question’s experience and what they’re going off of.

[danaris]

As a counterexample to your assertion, I've seen it a lot on both sides of the RTO discourse.

[freehorse]

Maybe you are not aware of such kinds of topics, but yes it is asked often. It is asked for stimulants, for microdosing psychedelics, for behavioural interventions or workplace policies/processses. Whenever there are any kind of productivity claims, it is asked, and it should be asked.

[tbrownaw]

It's not that hard to review how much you actually got done and check whether it matches how much it felt like you were getting done.

[freehorse]

To do that properly, one needs some kind of control, which is hard to do with one person. It should be doable with proper effort, but far from trivial, because it is not enough to measure what you actually did in one condition, you have to compare it with sth. And then there can be a lot of noise for n=1: when you use LLMs, maybe you happen to have to solve harder tasks. So you need at least to do it over quite a lot of time, or make sure the difficulty of tasks is similar. If you have a group of people, you can put them into groups instead and thus not care as much for these parameters, because you can assume that when you average this "noise" will cancel out.

[fragmede]

The problem isn't a delta between what got done and how much it felt like got done. The problem is it's not known how it would have taken you to do what got done unless you do it twice. Once by hand and once with an LLM, and then compare. Unfortunately, regardless of what you find, HN will be rushing to say N=1, so there's little incentive to report on any individual results.

[emp17344]

In fact, when this was studied, it was found that using AI actually makes developers less productive:

https://metr.org/blog/2025-07-10-early-2025-ai-experienced-o...

[coderjames]

> This friend told me she can't work without ChatGPT anymore.

It doesn't say she chooses to use it; it says she can't work without using it. At my workplace, senior leadership has mandated that software engineers use our internal AI chat tooling daily, they monitor the usage statistics, and are updating engineering leveling guides to include sufficient usage of AI being required for promotions. So I can't work without AI anymore, but it doesn't mean I choose to.

[echelon]

Serious thought.

What if people are using LLMs to achieve the same productivity with more cost to the business and less time spent working?

This, to me, feels incredibly plausible.

Get an email? ChatGPT the response. Relax and browse socials for an hour. Repeat.

"My boss thinks I'm using AI to be more productive. In reality, I'm using our ChatGPT subscription to slack off."

That three day report still takes three days, wink wink.

AI can be a tool for 10xers to go 12x, but more likely it's also that AI is the best slack off tool for slackers to go from 0.5x to 0.1x.

And the businesses with AI mandates for employees probably have no idea.

Anecdotally, I've seen it happen to good engineers. Good code turning into flocks of seagulls, stacks of scope 10-deep, variables that go nowhere. Tell me you've seen it too.

[johnnyanmac]

Yeah, I think this is why it's more important to shift the question to "is the team/business more productive". If a 0.5xer manager is pushing 0.1x work and a 1xer teammate needs to become a 1.5xer to fix the slop, then we have this phenomenon where the manager can feel way more productive, while the team under him is spending more time just to fix or throw out his slop.

Both their perspectives are technically right. But we'll either have burned out workers or a lagging schedule as a result in the long term. I miss when we thought more long term about projects.

[mrbungie]

That's Jevons paradox for you.

[y0eswddl]

There's literally a study out the shows when developers think LLMs are making them 20% faster, it turned out to be making them 20% less productive:

https://arxiv.org/abs/2507.09089

[spicyusername]

I mean... there are many situations in life where people are bad judges of the facts. Dating, finances, health, etc, etc, etc.

It's not that hard to imagine that your friend feels more productive than she actually is. I'm not saying it's true, but it's plausible. The anecdata coming out of programming is mostly that people are only more productive in certain narrow use cases and much less productive in everything else, relative to just doing the work themselves with their sleeves rolled up.

But man to seeing all that code gets spit out on the screen FEEL amazing, even if I'm going to spend the next few hours needing to edit it, for the next few months managing the technical debt I didn't notice when I merged it.

[mrbungie]

> What kind of question is that? Seriously. Are some people here so naive to think that tens of millions out there don’t know when something they choose to use repeatedly multiple times a day every day is making their life harder?

That's just an appeal to masses / bandwagon fallacy.

> Is it so hard to believe that ChatGPT is actually productive?

We need data, not beliefs and current data is conflicting. ffs.

[bakugo]

You're working under the assumption that punching a prompt into ChatGPT and getting up to grab some coffee while it spits out thousands of tokens of meaningless slop to be used as a substitute for something that you previously would've written yourself is a net upgrade for everyone involved. It's not. I can use ChatGPT to write 20 paragraph email replies that would've previously been a single manually written paragraph, but that doesn't mean I'm 20x more productive.

And yes, ChatGPT is kinda like an addictive drug here. If someone "can't work without ChatGPT anymore", they're addicted and have lost the ability to work on their own as a result.

[KoolKat23]

That's a very broad assumption.

It's no different to a manager that delegates, are they less of a manager because they entrust the work to someone else? No. So long as they do quality checks and take responsibility for the results, wheres the issue?

Work hard versus work smart. Busywork cuts both ways.

[aurareturn]

You’re assuming that there is zero quality check and that managers and clients will accept anything chatgpt generates.

Let’s be serious here. These are still professionals and they have a reputation. The few cases you hear online of AI slop in professional settings is the exception. Not the norm.

[isamuel]

> And yes, ChatGPT is kinda like an addictive drug here. If someone "can't work without ChatGPT anymore", they're addicted and have lost the ability to work on their own as a result.

Come on, you can’t mean this in any kind of robust way. I can’t get my job done without a computer; am I an “addict” who has “lost the ability to work on my own?” Every tool tends to engender dependence, roughly in proportion to how much easier it makes the life of the user. That’s not a bad thing.

[johnnyanmac]

> you can’t mean this in any kind of robust way.

Why not?

>I can’t get my job done without a computer; am I an “addict” who has “lost the ability to work on my own?”

It's very possible. I know people love bescmirching the "you won't always have a calculator" mentality. But if you're using a calculator for 2nd grade mental math, you may have degregaded too far. It varies on the task, of course.

>Every tool tends to engender dependence, roughly in proportion to how much easier it makes the life of the user. That’s not a bad thing.

Depends on how it's making it easier. Phones are an excellent example. They make communication much easier and long distance communication possible. But if it gets to the point where you're texting someone in the next room instead of opening your door, you might be losing a piece of you somewhere.

[danaris]

There's a big difference between needing a tool to do a job that only that tool can do, and needing a crutch to do something without using your own faculties.

LLMs are nothing like a computer for a programmer, or a saw for a carpenter. In the very best case, from what their biggest proponents have said, they can let you do more of what you already do with less effort.

If someone has used them enough that they can no longer work without them, it's not because they're just that indispensable: it's because that someone has let their natural faculties atrophy through disuse.

[bakugo]

> I can’t get my job done without a computer

Are you really comparing an LLM to a computer? Really? There are many jobs today that quite literally would not exist at all without computers. It's in no way comparable.

You use ChatGPT to do the things you were already doing faster and with less effort, at the cost of quality. You don't use it to do things you couldn't do at all before.

[johnnyanmac]

>Is it so hard to believe that ChatGPT is actually productive?

Given what I've seen in the educational sector: yes. Very hard. We already had this massive split in extremes between the highly educated and the ones who struggle. The last thing we need is to outsource the aspect of thinking to a billionaire tech company.

The slop you see in the workplace isn't encouraging either.

[Etheryte]

The recent MIT report on the state of AI in business feels relevant here [0]:

> Despite $30–40 billion in enterprise investment into GenAI, this report uncovers a surprising result in that 95% of organizations are getting zero return.

There's no doubt that you'll find anecdotal evidence both for and against in all variations, what's much more interesting than anecdotes is the aggregate.

[0] https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Bus...

[sigbottle]

I think it's true that AI does deliver real value. It's helped me understand domains quickly, be a better google search, given me code snippets and found obscure bugs, etc. In that regard, it's a positive on the world.

I also think it's true that AI is nowhere near AGI level. It's definitely not currently capable of doing my job, not by a long shot.

I also think that, throwing trillions of dollars at AI for a "a better google search, code snippet generator, and obscure bug finder" is a contentious question, and a lot of people oppose it for that reason.

I personally still think it's kind of crazy we have a technology to do things that we didn't have just ~2 years before, even if it just stagnates right here. Still going to be using it every day, even if I admittedly hate a lot of parts of it (for example, "thinking models" get stuck in local minima way too quickly).

At the same time, don't know if it's worth trillions of dollars, at least right now.

So all claims on this thread can be very much true at the same time, just depends on your perspective.

[johnnyanmac]

I have my criticisms of LLM's, but anyone in 2025 trying to sell you AGI is selling you a bridge made of snake oil. The aspect of the job market won't even be the biggest question the day we truly achieve AGI.

>At the same time, don't know if it's worth trillions of dollars, at least right now.

The revenue numbers sure don't think so. And I don't think this economy can support "trillions" of spending even if it wanted to. That's why the bubble will pop, IMO.

[jt2190]

That report also mentions individual employees using their own personal subscriptions for work, and points to it as a good model for organizations to use when rolling out the tech (i.e. just make the tools available and encourage/teach staff how they work). That sure doesn’t make it sound like “zero return” is a permanent state.

[Workaccount2]

Ah yes, the study that everyone posts but nobody reads

>Behind the disappointing enterprise deployment numbers lies a surprising reality: AI is already transforming work, just not through official channels. Our research uncovered a thriving "shadow AI economy" where employees use personal ChatGPT accounts, Claude subscriptions, and other consumer tools to automate significant portions of their jobs, often without IT knowledge or approval.

>The scale is remarkable. While only 40% of companies say they purchased an official LLM subscription, workers from over 90% of the companies we surveyed reported regular use of personal AI tools for work tasks. In fact, almost every single person used an LLM in some form for their work.

[johnnyanmac]

IT nightmares aside, this only makes the issue worse, if +it is so widespread to the point where some are sneaking to use it personally, and they still can't make a business more productive/profitable: well, that bubble is awfully wobbly

[crazygringo]

No. The aggregate is useless. What matters is the 5% that have positive return.

In the first few years of any new technology, most people investing it lose money because the transition and experimentation costs are higher than the initial returns.

But as time goes on, best practices emerge, investments get paid off, and steady profits emerge.

[Gerardo1]

On the provider end, yes. Not on the consumer end.

These are business customers buying a consumer-facing product.

[crazygringo]

No, on the consumer end. The whole point is that the 5% profitable is going to turn to 10%, 25%, 50%, 75% as companies work out how to use AI profitably.

It always takes time to figure out how to profitably utilize any technological improvement and pay off the upfront costs. This is no exception.

[johnnyanmac]

Can we both at least agree that 95% of comapnies investing and failing in a technology with 400b+ dollars of investment constitutes a bubble popping? I pretty much agree with you otherwise and that is what the article comes down to as well:

>I believe both sides are right. Like the 19th century railroads and the 20th century broadband Internet build-out, AI will rise first, crash second, and eventually change the world.

[admissionsguy]

When your work consists of writing stuff disconnected from reality it surely helps to have it written automatically.

[patapong]

On the other hand, it's a hundreds-of-billions of dollars market...

[Gerardo1]

What is?

[NoGravitas]

Writing stuff disconnected from reality, I assume.

[nizsle]

Greater output doesn't always equal greater productivity. In my days in the investing business we would have junior investment professionals putting together elaborate and detailed investment committee memos. When it came time to review a deal in the investment committee meetings we spent all our time trying to sift through the content of the memos and diligence done to date to identify the key risks and opportunities, with what felt like a 1:100 signal to noise ratio being typical. The productive element of the investment process was identifying the signal, not producing the content that too often buries the signal deeper. Imo, AI tools to date make it so much easier to create content which makes it harder to be productive.

[thisisit]

> This friend told me she can't work without ChatGPT anymore

I am curious what kind of work is she using ChatGPT such that she cannot do without it?

> ChatGPT released data showing that programming is just a tiny fraction of queries people do

People are using it as search engine, getting dating advice and everything under the sun. That doesn't mean there is business value - so to speak. If these people had to pay say $20 a month for this access, are they willing to do so?

The poster's point was that coding is an area which is paying consistently for LLM models so much that every model has a coding specific version. But we don't see same sort of specialized models for other areas and the adoption is low to nonexistent.

[FridgeSeal]

> what kind of work is she using ChatGPT such that she cannot do without it?

Given they said this person worked at PwC, I’m assuming it’s pointless generic consultant-slop.

Concretely it’s probably godawful slide decks.

[johnnyanmac]

>Where does this notion that LLMs have no value outside of programming come from?

Well this article cites 400b of spending for 12b of revenue. That's not zero value, but it definitely showing overvalue. We're not paying that level of money back with consumer level goods.

Now is B2B valuable? Maybe. But it's really tough valuating that with how businesses are operating c. 2025.

> ChatGPT released data showing that programming is just a tiny fraction of queries people do.

yes, but it's not 2010 anymore. Companies are already on ChatGPT's neck trying to get RoI's. They can't run insolvent for a decade at this level of spending like all the FAANG's did in yestr-decade.

[bakugo]

> This friend told me she can't work without ChatGPT anymore.

This isn't a sign that ChatGPT has value as much as it is a sign that this person's work doesn't have value.

[aurareturn]

What kind of logic is this?

ChatGPT automates much of my friend's work at PwC making her more productive --> not a sign that ChatGPT has any value

Farming machines automated much of what a farmer used to have to do by himself making him more productive --> not a sign that farming machines have any value

[dsr_]

The output of a farm is food or commodities to be turned into food.

The output of PwC -- whoops, here goes any chance of me working there -- is presentations and reports.

“We’re entering a bold new chapter driven by sharper thinking, deeper expertise and an unwavering focus on what’s next. We’re not here just to help clients keep pace, we’re here to bring them to the leading edge.”

That's on the front page of their website, describing what PwC does.

Now, what did PwC used to do? Accounting and auditing. Worthwhile things, but adjuncts to running a business properly, rather than producing goods and services.

[aurareturn]

The output of her work isn’t presentations and reports. The actual output is raising money and making successful deals. This requires convincing investors mostly which is very hard to do.

Look up what M&A is.

[johnnyanmac]

>Look how what M&A is.

Mergers and Aquisitions? If that's the right acronym I hate it even more, thank you.

But yes, I can see how automating the BS of corporate culture then using it to impress people (who also don't care anyway) by saying "I made this with AI" can be "productive". Not really a job I can do, though.

[deadbabe]

I find it’s mostly a sign of how lazy people get once you introduce them to some new technology that requires less effort for them.

[manishsharan]

Most developers can't do much work without an IDE and Chrome + Google.

Would you say that their work has no value?

[johnnyanmac]

This is probably the only place I can properly say "Programmers should be brought up with vim and man pages", so I'll say it here.

Anyways, IDE's don't try to offload the thinking for you, it's more like an abacus. You still need to work in it a while and learn the workflow before it's more efficient than a text editor + docs.

Chrome is a trickier aspect, because the reality is that a lot of modern docs completely suck. So you rely less on official documentation and more about how others have navigated an IDE and if those options work for you. I'd rather we make proper documentation than offload it into a black box that may or may not understand what it's spouting out to you, though.

[lowsong]

This says more about PwC and what M&A people do all day than it does about ChatGPT.

[kumarm]

[even in programming it's inconclusive as to how much better/worse it makes programmers.]

Try building something new in claude code (or codex etc) using a programming language you have not used before. Your opinion might change drastically.

Current AI tools may not beat the best programmer, they definitely improves average programmer efficiency.

[piker]

> Try building something new in claude code (or codex etc) using a programming language you have not used before. Your opinion might change drastically.

Try changing something old in claude code (or codex etc) using a programming language you have used before. Your opinion might change drastically.

[Wowfunhappy]

I have! Claude is great at taking a large open source project and adding some idiosyncratic feature I need.

[UrineSqueegee]

this is literally how i maintain the code at my current position, if i didn't have copilot+ i would be cooked

[aurareturn]

Same here. Coworker left. I now maintain a bunch of Go code and I had never written any before.

I use copilot in agent mode.

[Gerardo1]

...what were you doing before?

That's bread and butter development work.

[UrineSqueegee]

didnt have to maintain any code before thankfully

[tymonPartyLate]

I did just that and I ended up horribly regretting it. The project had to be coded in Rust, which I kind of understand but never worked with. Drunk on AI hype, I gave it step by step tasks and watched it produce the code. The first warning sign was that the code never compiled at the first attempt, but I ignored this, being mesmerized by the magic of the experience. Long story short, it gave me quick initial results despite my language handicap. But the project quickly turned into an overly complex, hard to navigate, brittle mess. I ended up reading the Rust in Action book and spending two weeks cleaning and simplifying the code. I had to learn how to configure the entire tool chain, understand various cargo deps and the ecosystem, setup ci/cd from scratch, .... There is no way around that.

It was Claude Code Opus 4.1 instead of Codex but IMO the differences are negligible.

[fabian2k]

AI can be quite impressive if the conditions are right for it. But it still fails at so many common things for me that I'm not sure if it's actually saving me time overall.

I just tried earlier today to get Copilot to make a simple refactor across ~30-40 files. Essentially changing one constructor parameter in all derived classes from a common base class and adding an import statement. In the end it managed ~80% of the job, but only after messing it up entirely first (waiting a few minutes), then asking again after 5 minutes of waiting if it really should do the thing and then missing a bunch of classes and randomly removing about 5 parenthesis from the files it edited.

Just one anecdote, but my experiences so far have been that the results vary dramatically and that AI is mostly useless in many of the situations I've tried to use it.

[manishsharan]

This is exactly my experience. We wanted to modernize a java codebase by removing java JNDI global variables. This is a simple though tedious task. And we tried Claude Code and Gemini. Both of these results were hilarious.

[Workaccount2]

LLMs are awful at tedious tasks. Usually because it involves massive context.

You will have much more success if you can compartmentalize and use new LLM instances as often as possible.

[benregenspan]

One thing I like for this type of refactoring scenario is asking it to write a codemod (which you can of course do yourself but there's a learning curve). Faster result that takes advantage of a deterministic tool.

[hemloc_io]

Yeah I've used it for personal projects and it's 50/50 for me.

Some of the stuff generated I can't believe is actually good to work with long term, and I wonder about the economics of it. It's fun to get something vaguely workable quickly though.

Things like deepwiki are useful too for open source work.

For me though the core problem I have with AI programming tools is that they're targeting a problem that doesn't really exist outside of startups, not writing enough code, instead of the real part of inefficiency in any reasonably sized org, coordination problems.

Of course if you tried to solve coordination problems, then it would probably be a lot harder to sell to management because we'd have to do some collective introspection as to where they come from.

[johnnyanmac]

>Of course if you tried to solve coordination problems, then it would probably be a lot harder to sell to management because we'd have to do some collective introspection as to where they come in.

Sad but true. Better to sell to management and tagline it as "you don't need a whole team anymore.", or going so far as "you can do this all by yourself now!".

Sadly managers usually have more money to spend than the workers too, so it's more profitable.

[randomNumber7]

> For me though the core problem I have with AI programming tools is that they're targeting a problem that doesn't really exist outside of startups

If you work in science it's great to have s.th. that spits out mediocre code for your experiments.

[tbrownaw]

> Try building something new in claude code (or codex etc) using a programming language you have not used before. Your opinion might change drastically.

So it looks best when the user isn't qualified to judge the quality of the results?

[thisisit]

> using a programming language you have not used before

haven't we established that if you are layman in an area AI can seem magical. Try doing something in your established area and you might get frustrated. It will give you the right answer with caveats - code which is too verbose, performance intensive or sometimes ignoring best security practices.

[philippta]

> they definitely improves average programmer efficiency

Do we really need more efficient average programmers? Are we in a shortage of average software?

[Cthulhu_]

Average programmers do not produce average software; the former implements code, the latter is the full picture and is more about what to build, not how to build it. You don't get a better "what to build" by having above-average developers.

Anyway we don't need more efficient average programmers, time-to-market is rarely down to coding speed / efficiency and more down to "what to build". I don't think AI will make "average" software development work faster or better, case in point being decades of improvements in languages, frameworks and tools that all intend to speed up this process.

[jennyholzer]

> Are we in a shortage of average software?

Yes. The "true" average software quality is far, far lower than the average person perceives it to be. ChatGPT and other LLM tools have contributed massively to lowering average software quality.

[layer8]

I don’t understand how your three sentences mesh with each other. In any case, making the development of average software more efficient doesn’t by itself change anything about its quality. You just get more of it faster. I do agree that average software quality isn’t great, though I wouldn’t attribute it to LLMs (yet).

[lowsong]

> using a programming language you have not used before

But why would I do that? Either I'm learning a new language in which case I want to be as hands-on as possible and the goal is to learn, not to produce. Or I want to produce something new in which case, obviously, I'd use a toolset I'm experienced in.

[jbstack]

There are plenty of scenarios where you want to work with a new language but you don't want to have to dedicate months/years of your life to becoming expert in it because you are only going to use it for a one-time project.

For example, perhaps I want to use a particular library which is only available in language X. Or maybe I'm writing an add-on for a piece of software that I use frequently. I don't necessarily want to become an expert in Elisp just to make a few tweaks to my Emacs setup, or in Javascript etc. to write a Firefox add-on. Or maybe I need to put up a quick website as a one-off but I know nothing about web technologies.

In none of these cases can I "use a toolset I'm experienced in" because that isn't available as an option, nor is it a worthwhile investment of time to become an expert in the toolset if I can avoid that.

[layer8]

How can I possibly assess the results in a programming language I haven’t used before? That’s almost the same as vibe coding.

[jbstack]

The same way you assess results in a programming language you have used before. In a more complicated project that might mean test suites. For a simple project (e.g. a Bash script) you might just run it and see if it does what you expect.

[layer8]

The way I assess results in a familiar programming language is by reviewing and reasoning through the code. Testing is necessary, but not sufficient by any means.

[edanm]

Out of curiosity, how do you assess software that you didn't write and just use, and that is closed source? Don't you just... use it? And see if it works?

Why this is inherently different?

[majewsky]

You are correct that this is indeed a mostly unsolved problem. In chapter 15 of "The Mythical Man-Month", Fred Brooks called for all program documentation to not only for how to use a program, but also for how to modify a program [1] and, relevant to this discussion, for how to believe a program. This was before automated tests and CI/CD were a thing, so he advocated for shipping testcases with the program that the user could review and execute at any time. It's now 50 years later, and this is one of the many lessons in that book that we've collectively not picked up on enough.

[1] Side-note: This was written at a time when selling software as a standalone product was not really a thing, so everything was open-source and the "how to modify" part was more about how to read and understand the code, e.g. architecture diagrams.

[piva00]

The question is: is that value worth the US$400bi per year of investment sucking out all the money from other ventures?

It's a damn good tool, I use it, I've learned the pitfalls, it has value but the inflation of potential value is, by definition, a bubble...

[player1234]

It is not worth it and it is not even that impressive considering the cost.

If you told me that you would spend half a trillion and the best minds on reading the whole internet, then with some statistical innovation try to guess the probable output of an input. The way it works now seems about right, probably a bit disappointing even.

I would also say, it seems cool and you could do that, but why would you? At least when the training is done it is cheap to use right? No!? What the actual fuck!

[raincole]

> adoption is low to nonexistent outside of programming

In the last few months, every single non-programmer friend I've met has ChatGPT installed on their phone (N>10).

Out of all the people that I know enough to ask if they have ChatGPT installed, there is only one who doesn't have it (my dad).

I don't know how many of them are paying customers though. IIRC one of them was using ChatGPT to translate academic writing so I assume he has pro.

[hermannj314]

My daughter and her friends have their own paid chatgpt. She said she uses it to help with math homework and described to me exactly why I bought a $200 TI-92 in the 90s with a CAS.

Adoption is high with young people.

[Gigachad]

I’ve been trying out locally run models on my phone. The iPhone 17 is able to run some pretty nice models, but they lack access to up to date information from the web like ChatGPT has. Wonder if some company like Kagi would offer some api to let your local model plug in and run searches.

[data-ottawa]

Kagi do but it’s fairly expensive (for my taste) and you have to email them for access.

There are other companies that provide these tools for anything supporting MCP.

[fxtentacle]

That would be perfect. Especially because Kagi could also return search results as JSON to the AI if they control both sides of the interaction.

[cornholio]

It's silly to say that the only objective that will vindicate AI investments is AGI.

Current batch of deep learning models are fundamentally a technology for labor automation. This is immensely useful in itself, without the need to do AGI. The Sora2 capabilities are absolutely wild (see a great example here of what non-professional users are already able to create with it: https://www.youtube.com/watch?v=HXp8_w3XzgU )

So only looking at video capabilities, or at coding capabilities, it's already ready to automate and upend industries worth trillions in the long run.

The emerging reasoning capabilities are very promising, able to generate new theories and make scientific experiments in easy to test fields, such as in vitro drug creation. It doesn't matter if the LLM hallucinates 90% of the time, if it correctly reasons a single time and it can create even a single new cancer drug that passes the test.

These are all examples of massive, massive economic disruption by automating intellectual labor, that don't require strict AGI capabilities.

[hemloc_io]

Regardless of my opinions on if you're correct about this, I'm not an ML expert so who knows, I'd be very happy if we cured cancer so I hope you're correct and the video is a cool demo.

I don't believe the risk vs reward on investing a trillion dollars+ is the same when your thesis changes from "We just need more data/compute and we can automate all white collar work"

to

"If we can build a bunch of simulations and automate testing of them using ML then maybe we can find new drugs" or "automate personalized entertainment"

The move to RL has specifically made me skeptical of the size of the buildout.

[rafaelmn]

If you calculate the investment into AI and then divide by say 100k that's how many man-years need to replace with AI to be cost effective as labor automation the numbers aren't that promising given the current level of capability.

[Gerardo1]

Don't even need to get too fancy with it. Open AI has publicly committed to ~$500B in spending over the next several years (nevermind even they don't expect to actually bring that much revenue in)

$500B/$100,000 is 5 million, or 167k 30-year careers.

The math is ludicrous, and the people saying it's fine are incomnprehensible to me.

Another comment on a similar post just said, no hyperbole, irony, or joke intended: "Just you switching away from Google is already justifying 1T infrastructure spend."

[cornholio]

I don't think I follow. 1 trillion total investment divided by 100k yields 10 million man years, or 300k man-careers.

Just the disruption we can already see in the software industry are easily of that magnitude.

[rafaelmn]

>Just the disruption we can already see in the software industry are easily of that magnitude.

WTF ? Where are you seeing that ?

Also no you can't calculate 100k over 30 years as 3M because you expect investment growth - lets say stock market average of 7 percent per year that investment must return like 24 million in 30 years otherwise its not worth it. That means 8 trillion in next 30 years if you look over that long of an investment period.

And who in the hell is going to capture 30 years of profit with model/compute investments made today.

The math only maths within short timeframes - hardware will get amortized in 5 years, model obsolete in even less. So best case scenario you have to displace 2 million people and capture their output to repay that. Not with future tech - with tech investments made today.

[cornholio]

The global employment in software development and adjacent is in the tens of millions. To say the impact of AI code automation will be, at max, a rounding error of just 1-2% of that is just silly; currently, the junior pipeline is almost frozen in the global north, entire batches of graduates can't find jobs in tech.

Sure, the financial math over 30 years does not follow elementary arithmetic, and if the development hits a wall tomorrow they will have trouble recovering the investment just from code automation tools.

But this is a clearly nonsense scenario, the tech is rapidly expanding to other fields that have obvious potential to automate. This is not a pie-in the sky future technology yet to be invented, it's obvious productization of latent capability, similar to the early internet days. There might be some overshoots but the latent potential is all there, the AI investments are looking to be the first movers in that enormously lucrative space and take, what seem to me, reasonable financial risks in light of the rewards.

My claim is not that AGI will soon be available, but that applying existing frontier models on the entire economy, in the form of mature, yet to be developed products, will easily generate disruption that has a present value in the trillions.

[rafaelmn]

You do understand that you don't replace a 100k developer and call it a day - you have to charge the same company 100k for your AI tools. No model is nowhere near close today - they are having trouble convincing enterprises to pay less than 100$ per employee. The current models do not math at all, the only way these investments work is if models get fundamentally better.

[maeln]

> Current batch of deep learning models are fundamentally a technology for labor automation. This is immensely useful in itself, without the need to do AGI. The Sora2 capabilities are absolutely wild (see a great example here of what non-professional users are already able to create with it: https://www.youtube.com/watch?v=HXp8_w3XzgU )

> So only looking at video capabilities, or at coding capabilities, it's already ready to automate and upend industries worth trillions in the long run.

Can Sora2 change the framing of a picture without changing the global scene ? Can it change the temperature of a specific light source ? Can it generate a 8k HDR footage suitable for re-framing and color grading ? Can it generate minute long video without loosing coherence ? Actually, can it generate a few seconds without having to reloop with the last frame and have these obnoxious cuts that the video you pointed has ? Can it reshoot the same exact scene with just one element altered ?

All the video models right now are only good at making short, low-res, barely post-processable video. The kind of stuff you see on social media. And considering the metrics on ai-generated video on social media right now, for the most part, nobody want to look at them. They might replace the bottom of the barrel of social media posting (hello cute puppy videos), but there is absolutely nothing indicating that they migth automate or upend any real industry (be used in the pipeline, yeah maybe, why not, automate ? Won't hold my breath).

And the argument of their future capabilities, well ... It's been 50+ years that we should have fusion in 20 years.

Btw, the same argument can be made for LLM and image-gen tech in any creative purposes. People severly underestimate just how much editing, re-work, purpose and pre-production steps are involved in any major creative endeavor. Most model are just severly ill suited for that work. They can be useful for some stuff (specificaly, for editing images, ai-driven image fill do work decently for exemple), but overall, as of right now, they are mostly good at making low quality content. Which is fine I guess, there is a market for it, but it was already a market that was not keen on spending money.

[data-ottawa]

This is very surface level criticism.

Qwen image and nano banana can both do that with images, there’s zero reason to think we can’t train video models for masking.

This feels a lot like critiquing stable diffusion over hands and text, which the new SOTA models all handle well.

One of the easiest iterations on these models is to add more training cases to the benchmarks. That’s a timeline of months, not comparable to forecasting progress over 20 years like fusion.

[maeln]

> This is very surface level criticism.

Is it now. I don't think being able to accurately and predictably make changes to a shot, a draft, a design is surface level in production.

> Qwen image and nano banana can both do that with images, there’s zero reason to think we can’t train video models for masking.

Tell them to change the tilt of the camera roughly 15 degree left without changing anything else in the scene and tell me if it works.

> This feels a lot like critiquing stable diffusion over hands and text, which the new SOTA models all handle well.

Well does a lot of heavy lifting there.

> One of the easiest iterations on these models is to add more training cases to the benchmarks. That’s a timeline of months, not comparable to forecasting progress over 20 years like fusion.

And what if the model itself is the limiting factor ? The entire tech ? Do we have any proof that in the future the current technologies might be able to handle the cases I spoke about ?

Also, one thing that I didn't mention in the first post. Assuming that the tech does come to the point I can be used to automate a lot of the production. If Throwing a few millions to buy a GPU cluster is enough to be able to "generate" a relatively high quality movie or series, the barrier to entry will be incredibly low. The cost will be driven down, the amount of production will be very high and overall it might not be a trillion dollar industry no more.

[____mr____]

> They might replace the bottom of the barrel of social media posting (hello cute puppy videos)

Lay off. Only respite I get from this hell world is cute Rottweiler videos

[asdf1280]

The problem is that it’s already commodified; there’s no moat. The general tech practice has been capture the market by burning vc money, then jack up prices to profit. All these companies are burning billions to generate a new model and users have already proven there is no brand loyalty. They just hop to the new one when it comes out. So no one can corner the market and when the VC money runs out they’ll have to jack up prices so much that they’ll kill their market

[Wowfunhappy]

> The problem is that it’s already commodified; there’s no moat.

From an economy-wide perspective, why does that matter?

> users have already proven there is no brand loyalty. They just hop to the new one when it comes out.

Great, that means there might be real competition! This generally keeps prices down, it doesn't push them up! It's true that VCs may end up unhappy, but will they be able to do anything about it?

[frogperson]

The most isnt with the llm crestors, its with nvidia, but even that is under seige by Chinese mskers.

[squidbeak]

Compute is the moat.

[jennyholzer]

You seem to be making an implicit claim that LLMs can create an effective cancer drug "10% of the time".

Smells like complete and total bullshit to me.

Edit: @eucyclos: I don't assume that Chat GPT and LLM tools have saved cancer researchers any time at all.

On the contrary, I assume that these tools have only made these critical researchers less productive, and made their internal communications more verbose and less effective.

[cornholio]

No, that's not the claim. The claim is that we will create a hypothetical LLM that, when tasked with a problem at the scientific frontier of molecular biology will, about 10% of the time, correctly reason about existing literature and reach conclusions that are valid or plausible to similar experts in the field.

Let's say you run that LLM one million times and get 100.000 valid reasoning chains. Let's say among them are variations on 1000 fundamentally new approaches and ideas, and out of those, you can actually synthesize in the laboratory 200 new candidate compounds, and out of those, 10 substance show strong in-vitro response, and then one of those completely cures some cancerous mice.

There you go, you have substantially automated the intellectual work of cancer research and you have one very promising compound you can start phase 1 trials that you didn't have before AI, and all without any AGI.

[eucyclos]

How many hours writing emails does it have to save human cancer researchers for it to be effectively true?

[onlyrealcuzzo]

> adoption is low to nonexistent outside of programming

Odd way to describe ChatGPT which has >1B users.

AI overviews have rolled out to ~3B users, Gemini has ~200M users, etc.

Adoption is far from low.

[TheCapeGreek]

> AI overviews have rolled out to ~3B users

Does that really count as adoption, when it has been introduced as a default feature?

[onlyrealcuzzo]

Yes, if people are interacting with them, which they are.

HN seems to think everyone is like in the bubble here, which thinks AI is completely useless and wants nothing to do with it.

Half the world is interacting with it on a regular basis already.

Are we anywhere near AGI? Probably not.

Does it matter? Probably not.

Inference costs are dropping like a rock, and usage is continuing to skyrocket.

[majewsky]

I am interacting with AI daily through Google products. YouTube is consistently giving me auto-translated titles that are either hilarious or wrong, and I desparately want to turn this bullshit off, but I can't, because it's not giving me an option.

That's the kind of adoption that should just be put up for adoption instead.

(And of course, the reason that I can tell that the auto-translated video titles are hilarious and/or wrong is because they are translating into a language that I speak from a language that I also speak, but apparently the YouTube app's dev team cannot fathom that a person might speak more than one language.)

[ACCount37]

Mostly agreed, but AI overviews are a very bad example. Google can just force feed its massive search user base whatever bullshit it damn pleases. Even if it has negative value to the users.

I don't actually think that AI overviews have "negative value" - they have their utility. There are cases where I stop my search right after reading the "AI overview". But "organic" adoption of ChatGPT or Claude or even Gemini and "forced" adoption of AI overviews are two different beasts.

[neutronicus]

My father (in his 70s) has started specifically looking for the AI overview, FWIW.

He has not engaged with any chatbot, but he thinks of himself as "using AI now" and thinks of it as a value-add.

[apercu]

"Where's the business value? Right now it doesn't really exist, adoption is low to nonexistent outside of programming and even in programming it's inconclusive as to how much better/worse it makes programmers."

The business model is it is data collection about you on steroids, and that the winning company will eclipse Meta in value.

It's just more ad tech with multipliers, and it will continue to control thought, sway policy and decide elections. Just like social media does today.

[mnky9800n]

i think it is more like maps. before 2004, before google maps, the way we interacted with the spatial distribution of places and things was different. all these ai dev tools like claude code as well as tools for writing, etc. are going to change the way we interact with our computers.

but on the other side, the reason everyone is so gung ho on all this is because these models basically allow for the true personalization of everything. They can build up enough context about you in every instance of you doing things online that they can craft the perfect ad experience to maximize engagement and conversion. that is why everyone is so obsessed with this stuff. they don't care about AGI, they care about maintaining the current status quo where a large chunk of the money made on the internet is done by delivering ads that will get people to buy stuff.

[UpsideDownRide]

I think there is a good flipside too. LLMs potentially enable generating custom made tooling tailored just for you. If you can get/provide data it's pretty easy to cook up solutions.

As an example - I'd never bother with mobile app just for myself since it's too annoying to get into for a somewhat small thing. Now I can chug along and have LLM fill in quickly my missing basic in the area.

[mnky9800n]

Yes. In my opinion the promise of LLMs isn’t whatever AGI robots doing everything for you. It’s a kind of holodeck or computer like in Star Trek. A tool that simply is able to translate your ideas into a computational representation.

[viking123]

I think there is real value, for instance nowadays I just use chatGPT as google replacement, brainstorming, and for coding stuff. It's quite useful and it would be hard to go back to time without this kind of tool. The 20 bucks a month is more than worth it.

Not sure though that do they make enough revenue and what will be the moat if more or less the best models will converge around the same level. For most normies, it might be hard to spot difference between gpt 5 and claude for instance. Okay for Grok the moat is that it doesn't pretend to be a pope and censor everything.

[skybrian]

Maybe you just haven’t heard of them? For example, just the other day I heard about a company using an LLM to provide advice to doctors. News to me.

https://www.prnewswire.com/news-releases/openevidence-the-fa...

> OpenEvidence is actively used across more than 10,000 hospitals and medical centers nationwide and by more than 40% of physicians in the United States who log in daily to make high-stakes clinical decisions at the point of care. OpenEvidence continues to grow by over 65,000 new verified U.S. clinician registrations each month. […] More than 100 million Americans this year will be treated by a doctor who used OpenEvidence.

More:

https://robertwachter.substack.com/p/medicines-ai-knowledge-...

[datadrivenangel]

I've had doctors google things in front of me. This may be an improvement.

[kalap_ur]

Likely not true re adoption. According to McKinsey November 2024 12% of employees in the US used AI for >30% of their daily tasks. I saw another research early this summer, it said that 40% of employees use AI. Adoption is already pretty relevant. The real question is: number of people x token requirement of their daily tasks equals how many tokens, and where are we there. Based on McK, we possibly around 17% unless remaining 50% of tasks requires just way more complexity, because then that would obviously mean the incremental tasks require maybe exponentially more tokens and then penetration will be indeed low. But for this we need to know total token need of daily tasks of average office worker.

[codingdave]

There is a middle ground where LLMs are used as a tool for specific use cases, but not applied universally to all problems. The high adoption of ChatGPT is the proof of this. General info, low accuracy requirements - perfect use case, and it shows.

The problem comes in when people then set expectations that a chat solution can solve non-chat problems. When people assume that generated content is the answer but haven't defined the problem.

We're not headed for AGI. We're also not going to just say, "oh, well, that was hype" and stop using LLMs. We are going to mature into an industry that understands when and where to apply the correct tools.

[aiisnotjustah]

I don't know what you mean by this at all tbh.

I don't think the researchers at the top think LLM is AGI.

DeepMind and co are already working on world models.

The biggest bottleneck right now is compute compute and compute. If an experiement takes MONTH to train, you want a lot more compute. You need compute to optimize what you already have like LLMs and then again a lot of compute to try out new things.

All of the compute/Datacenters and GPUs are not LLM GPUs. They are ML capable GPUs.

[tim333]

>the required mention that AI, specifically an LLM based approach to AGI, is important...

I don't think that's true. The people who think AI is important call it AI. The skeptics call it LLMs so they can say LLMs won't work. It's kind of a strawman argument really.

[kalleboo]

You can buy a rice cooker that claims it uses "AI", it's too much of a marketing buzzword to be useful.

[billy99k]

"Where's the business value? "

Have you ever used an LLM? I use it every day to help me with research and completing technical reports (which used to be a lot more of my time).

Of course you can't just use it blindly, but it definitely adds value.

[lm28469]

Does it bring more value than it cost ? That's the real question.

Nobody doubt it works, everybody doubt Altboy when he asks $7 trillion

[UpsideDownRide]

This question is pretty hard to answer without knowing the actual costs.

Current offerings are usually worth more than they cost. But since the prices are not really reflective of the costs it gets pretty muddy if it is a value add or not.

[player1234]

How did you measure this "a lot more of my time", please share your results and methodology!

[sakesun]

Have you read the article ? The cost is currently not justified for the benefit.

[joules77]

Cause the story is no more about Business or Economics. This is more like the nuke arms race in the 1940s. Red Queen Dynamics.

[calmoo]

FWIW Derek Thompson (the author of this blogpost) isn't exactly a 'business guy'

[jennyholzer]

If I'm not mistaken he's working with Ezra Klein to push the Democrats to embrace racism instead of popular economic measures.

Edit: I expect that these guys will try to make a J.D. Vance style Republican pivot in the next 4-8 years.

Second Edit:

Ezra Klein's recent interview with Ta-Nehisi Coates is very specifically why I expect he will pivot to being a Republican in the near future.

Listen closely. Ezra Klein will not under any circumstances utter the words "Black People".

Again and again, Coates brings up issues that Black People face in America, and Klein diverts by pretending that Coates is talking about Marginalized Groups in general or Trans People in particular.

Klein's political movement is about eradicating discussion of racial discrimination from the Democratic party.

Third Edit:

@calmoo: I think you're not listening to the nuances of my opinion, and instead having an intense emotional reaction to my well-justified claims of racism.

[Wowfunhappy]

We're very off topic, but if you're truly interested in Ezra Klein's worldview, I highly recommend his recent interview with Ta-Nehisi Coates. At minimum, I think you'll discover that Ezra's feelings are a lot more nuanced than you're making them out to be.

https://www.nytimes.com/2025/09/28/opinion/ezra-klein-podcas...

[calmoo]

I don't really want to discuss politics off the bat of my purely 'for your information' comment, but I think you're grossly misrepresenting Ezra Klein's worldview and not listening to the nuances of his opinion, and instead having an intense emotional reaction to his words. Take a step back and try to think a bit more rationally here.

Also your prediction of them making a JD vance republican pivot is extremely misguided. I would happily bet my life savings against that prediction.

[Cthulhu_]

Why would we want AGI? I've yet to read a convincing argument in favor (but granted, I never looked into it, I'm still at science-fiction doomerism). One thing that irks me is that people see it as inevitable, and that we have to pursue AGI because if we don't, someone else will. Or more bleak, if we don't actively pursue us, our malignant future AGI overlords will punish us for not bringing it into existence (roko's basilisk, the thing Musk and Grimes apparently bonded over because they're weird)