This article is probably not accurate. AI allows me to create hugely complex apps. Apps that I mostly don't understand the underlying code but can evaluate that it works fairly quick and effectively.

That said, in the corporate world, it's not that easy. There are tons of hoops to jump through and context switching all day long. So while my code is 100% AI generated these days, and I can make extremely complicated apps quickly at home, at work however, I'm burned out and completely checked out for the most part, entirely due to AI.

We don't have the same capabilities to burn tokens in the corporate world like we do at home. We don't have the creative freedoms we have at home. AI productivity is just not easily measured.

> This article is probably not accurate. AI allows me to create hugely complex apps.

To prove the article wrong even within the context of AI coding (which wasnt the focus of the article) you would have to present proof that creating these new AI apps resulted in you making lots of new money.

You'll have to take my word for it because I'm not going to disclose my private business or personal financial records.

But, I'm running a small, two person business and we are being paid by a large company to develop a robotics project. We're working at a very fast velocity and have fielded a prototype within a few months. I've been doing this kind of work for years, and we're more productive than larger teams (8-10 developers) I've employed before. Five years ago, I would have needed at least 4 senior engineers to do the work we're doing now, and we're moving faster than we could even then.

And I'm being compensated at a flat rate by a major company so we're making really good money. The customer is happy with the results. Claude easily writes 90% of the code we use.

There's other word for such code and it sounds similar. It's crappy. I was tempted to use Claude couple of days ago to help me with optimization and from 4Gb RAM and 21s it managed to squeeze 1Tb and 41s. When I asked to fixed it AI reduced it to 1Gb and 10s. It just removed most heavy business logic. So yeah. 3 days and this was only total waste of time and cash

How do you know it’s crappy? Have you read the code?

Your comment is strange, he clearly did in this context - that's why he knew what went wrong?

Fwiw, I currently consider code generated by models below opus level borderline unusable. It really is horrendous there.

With opus, it's still not good - but it's a lot better then some of my colleagues write.

Well written code, carefully planned out and thought through is still only possible by either hand writing or holding the hand of the model so much that you may as well just write by have, as you'll be quicker.

However, the code opus produces is good enough... So I rarely take action anymore in private projects

In which world throwing business logic to "optimize memory usage" isn't crappy solution? Or making 1T from 4G? It didn't returned decent solution, contrary it was worst than terrible

Guess who will be held responsible when your vibe-coded robot maims or kills someone.

I have enough experience writing code by hand and designing complex systems with a team working for me. In a sense, this is no different. When I had a half-dozen mid to senior level developers, I did not verify every line of code they wrote. I did not have any expectation that they would write perfect code, and they didn't.

But as the owner of the company, I was still liable for the product my employees made. That is the same thing here. It's not negligent to have Claude code write code for you at 10x the speed you can do it yourself. If I hadn't supervised my employees and put practices in place to control quality and safety, I would be liable. I'm actually far less worried about liability now, because I have my hands in every part of the system.

Calling it vibe coding is pejorative at this point, meant to imply that someone who has no skill or craft in software development just types a few prompts and gets something they didn't have a hand in the design or development. That's not what the professionals who are using AI coding are doing.

This is the crucial point.

It has vastly increased my hobby programming "velocity", but has improved my day job performance by perhaps a quarter at most; so much in programming is not the task of programming, but mapping the problem and dealing with the outside world.

I will not be providing proof here, but I went from someone who last wrote production code >15 years ago, to last week, a Fortune 50 demo client saying "this is far better than our internal tool." I am a solo builder.

I feel like I am an outlier, but I do exist.

Did you make money from it?

I am thrilled to report that after 1.5yrs of rice and beans: yes!

The first client has paid, and signed a 3 year contract. Not huge money, but 5 figures per year. That client didn't even see what the F50 demo user saw.

On personal projects AI has multiplied my velocity by about 10x.

In a corporate setting I've seen it reduce velocity by 80% in some cases.

https://news.ycombinator.com/item?id=48777257#48778335

I have been an AI sceptic for a long time. But we now throw in a native Android and native iOS app with apps we work on for clients now, because the amount of time saved makes it possible.

I could not move this fast on a corporate env though. I mean, I’m using half a dozen Chinese models, a Google AI pro sub, an OpenAI pro sub, and probably will add Claude to the list soon. We have also completely changed our workflows to align with how agentic programming works, including all the way to product management on the first place.

This article isn't focused on LLM for code generation.

Who is "making apps" at work? I keep seeing AI enthusiasts talk about how many apps they are making. I've never seen a real life dev whose job is churning out apps.

I think for some people "apps" just means any software.

I used to work with a guy who at some point had heard the term "macro" in regards to Word or Excel and at that point anything to do with computers was a "macro."

We'd get calls from him at the help desk saying "my mouse macros are not working" meaning i.e. that his mouse wasn't working, in that era because the roller ball was dirty.

Here's what happens: person comes up with an idea, asks Claude to generate it, gasps in awe, tries out for a few minutes, moves on because the idea wasn't that interesting.

When producing code is basically a solved problem, generating worthless amounts of code is literally one click away. Execution remains key, and still means iterating upon an idea for a long time, rather than generating a prototype with Claude and calling it a day.

Honestly for some it's just a compulsion that they can later brag about on forums, but it never results in anything worth showing the world.

The article could be accurate but misleading.

It focuses on the average (mean), but does not talk about the distribution at all. It groups ALL uses of AI together, which isn't very helpful in determining if it is worth it or not to use AI for a specific purpose.

For example, these stats could mean every single use gets 3% time savings, or some uses get 80% time savings and other use cases actually take more time.

This has always caught up to me in a bad way.

All it takes is some brittleness to need to take apart the app code and understand it. AI fixes one thing, breaks something completely unrelated.

And usually it’s full of lots of spaghetti code slop. Then I need to find ways to modularize it, make it testable, and at least at some level clean it up.

This is why I say AI sw eng is basically just working with legacy code. It has always turned into unraveling and refactoring the ball of mud to be sane for me and agents to continue working with.

More and more though I just go slower. Write much of the code myself. And setup good validation for the specific parts I trust an agent to work autonomously. I try to expand the surface of things that can be done well autonomously without losing my own grip on the code.

> Apps that I mostly don't understand the underlying code

So you’re just flinging tokens at the wall until something sticks. That’s not engineering.

Honestly, more like being a CEO / CTO that manages a ton of agents and these dumb fucks go off the rails at times and have to be yelled at. It's like herding cats. I've created (AI) entire harness systems, Electron apps, mobile app, SaaS products, video games, MCP servers and more.

Many of these things seem like they have the possibility of being their own products, but honestly, I'm just tired and burned out. I'm mostly building this stuff to remain relevant and prepared to pivot when the inevitable layoff comes. This layoff will come, for the same reason I'm able to create all this shit. CTO's and CEO's think this stuff will really work.

What the C-Suite doesn't know is... this is a fucking disaster waiting to happen. At some point the realization will come, that you have no moat. You have no unique ideas. If you do, they are easily copied. You're also giving over your IP to a company that is losing a fuck ton of money and will keep raising their prices. Worst of all, nobody including yourself will understand what your code is doing or forgetting to do. I foresee a bleak future.

I've had a decent amount of success in this industry. My plan is to ride this shit storm and see where it takes me, or do my best to help bring the entire system crashing down. By that, I mean, I'll release / contribute free and open source software, which mimics every business model that doesn't have a legal or financial moat. For instance... Slack, Miro, Shopify, the million AI wrapper products that are springing up every month etc.

You be surprise if people didnt do that

I’m not surprised. Of course people take shortcuts. They did it before LLMs came around.

If you don’t understand the output you don’t get to judge the quality.

> This article is probably not accurate. AI allows me to create hugely complex apps.

> That said, in the corporate world, it's not that easy.

The article is accurate, and describes this very paradox.

> Apps that I mostly don't understand the underlying code but can evaluate that it works fairly quick and effectively.

This is a contradiction. If you don't understand the code, you cannot in fact evaluate that it works correctly. If that's the approach you are taking, you are doing sloppy work that is going to blow up in your face sooner or later.

I agree, but some (most?) software being written doesn’t require a deep understanding to verify because the domain is small enough or you’re not required to solve for all of its intricacies. E.g. prototypes, internal tools, low scale CRUD apps, personal projects, etc.

I believe this is where the huge divide in perceived AI productivity in SW comes from. It’s folks working on low-understanding-required domains talking to folks working on high-understanding-required domains.

> E.g. prototypes, internal tools, low scale CRUD apps, personal projects, etc.

Those things shouldn't be hugely complex. They sound actually very simple.

> E.g. prototypes, internal tools, low scale CRUD apps, personal projects, etc.

My experience has been that small-ish projects like these are the most likely to contain code with bond-villain level of complexity (and success).

More than once have I been stuck going on Da Vinci Code-esque adventures to uncover bugs in prototypes years after the fact because the business pivoted away from what it was trying to do and later decided to pivot back, only to discover that despite the prototype and systems/libraries it worked with not changing, somehow it mysteriously doesn't work at all or it fails in convoluted ways despite being perfectly functional when it was originally shelved.

Read the first paragraph could tell it was AI written. Maybe if it was hand edited to be less AI I would have read more but it looked like the laziest Claude prose.

That'll be a no from me dawg.

Doesn't read like it's AI-written to me.

there's plenty of tells, e.g. "how to actually bank the part that is real" — AI tends to just "throw words out there" with very little information density

you COULD create a writer/reviewer loop hook to catch and rewrite sentences like this, but almost no one does

the result is usually very long essays that say very little (kind of like in grade school where we're graded on "length" so we just add words as filler)

I'm 50:50 on it. It's either AI written or written by someone who writes like an LLM does.

It's been heavily edited to reduce its AI-ness. Some tells remain, for example "Not X. It's Y."

Enough smell has been removed where it's tolerable.

I think we have to accept that most/all prose in the future will either be AI written or AI edited at least.

Fine so long as the content checks out (!important). I'm not reading it for its artistic or human value.

A year ago AI wrote roughly 0% of my code and now it writes roughly 100%.

Which is to say any AI study from a year ago is fairly out of date with the speed of advancement.

> A year ago AI wrote roughly 0% of my code and now it writes roughly 100%.

Did the amount of money you or your employer make rise by anywhere near 100% as well?

Because that is really the primary question behind articles like this focusing on the economic benefits.

But that itself doesn't tell us yet how much time it saved you at the end of the day.

As a developer, on my very best days I probably spent no more than 50% of my day actually writing code. So call it 4 hours of actual hands-on-keyboard coding.

But AI can write code much faster than a human. In 4 hours it might be able to write what would have taken me a week. Or more. Assuming it had proper specifications for that volume of code.

Do you still review the code it writes?

Typically it takes me as long to review code as it takes me to write it myself. The exception is repetitive coding tasks that cause my attention to drift, which LLMs can do very quickly.

The study is two months old (March 13, 2026).

Edit: Also, it's about writing not coding and it's point isn't that time isn't saved but even with time saved at the task, you don't get a broad decrease in total worked time for many/most workers.

The study was published in May 2025, and then revised in March 2026. It's based on data through December 2024, so virtually useless for saying anything about today.

Given that, I actually think the conclusions of the article are backwards. If the gpt-4/gpt-4o era had a measurable 3% improvement in productivity, how much more improvement are we getting from models today that are way way better?

The study is "new", but the data they are using is old. It has references GPT-4 for god sake.

Right but it looks back well before that.

AI hasn't gotten better, you've just given in to the hype.

Saving time does not automatically translate into higher productivity, or even lower costs. That should be obvious?

In fact, I would argue the that with AI, companies should expect to spend more on average, without necessarily seeing any meaningful cost savings nor increase in profits.

That does not mean they can escape this though. It is just like paying for ads, backlinks etc.

In what sense is AI like paying for ads? In one case you legitimately lose to the competition if you stop (prisoner's dilemma). In other case, whether and how much you lose depends almost entirely on how much it actually impacts your productivity in practice (or not).

If you're referring to the public perception benefits (supposing those even exist, which isn't clear to me), then it seems easy to make a lot of noise via PR while doing the minimal amount internally to explore the use case and not actually push it as hard as you say.

If we assume everyone has access to the same AI tools and those tools do help improve output a lot then you get the same prisoner's dilemma as ads.

The people using the AI tools to improve their output will set higher consumer expectations for their product or service. If you do not also use AI, you will not be able to keep up, and so you will lose business over time.

So you have to pay for AI just to keep up with the other people who pay for AI.

But then if everyone is paying for and using AI to improve their product or service, no one can steal market share from each other. So you are all now paying a lot and AI is doing a lot and the buyers are getting better results, but that does not necessarily lead to better financial results in terms of market share, revenue, costs or profits for the business that is now using AI

If AI works the way you suggest then we've literally been here before and we know exactly how this goes.

When tractors came along, farmers became dramatically more productive. Were the farmers who did things the old way "forced" to buy tractors in order to stay afloat? Over time, sure. But this was in no way zero-sum. More products made it to market (literally), more of the workforce was able to shift away from manual labor, and society became much better off overall. The people who moved to cities made dramatically more money, and the farmers who remained made more money too.

Edit: this is literally on the front page right now: https://news.ycombinator.com/item?id=48775979

This is not a prisoner's dilemma in any meaningful sense, unless you just like being inefficient and wasting a bunch of human effort.

(I'm not convinced yet whether AI actually works this way or not. I'm just saying, if it works this way, the economic theory is well-developed and we can predict fairly accurately how it's going to play out.)

The article gives valid advice in using AI on tasks AI is good at, but the article could be clearer on is what AI is good at and what AI is not so good at.

This is a 2024 survey, so it predates Claude Code and is mostly measuring GPT-4o:

https://bfi.uchicago.edu/wp-content/uploads/2025/04/BFI_WP_2...

AI is a multiplier, not an optimizer.

Engineers, (and many other types of specialists/professionals) can use it to speed up their work and increase output (even increase quality in some cases if done right).

But it's not gonna make inherently inefficient, political, and corrupted internal/company processes any more efficient, it might actually multiply those existing inefficiencies.

This article looks AI-written but I doubt it increased anyone’s paycheck.

I wouldn’t expect productivity to increase paychecks all that quickly except in special cases like that memory factory in South Korea. Higher productivity sometimes eventually results in more revenue and some companies competing for better workers by offering higher salaries (as has happened for software engineers), but this takes time and depends on companies believing that hiring better workers benefits them.

We aren’t really seeing that yet. We are seeing layoffs and companies being cautious about hiring. If it happens at all, it will take time.

Code isn't the bottleneck. Wait till devs find out making money is actually more work than just producing "systems".

>And in coding specifically, the speedup is real and comes with its own bill.

Oh so this article isn’t really about where AI spending is actually happening.

I HOPE people aren’t spending money on AI writing emails. That’s definitely not worth it.

Companies are spending the most on coding tools, not email writing. That’s the main value proposition right now.

It also doesn’t get into media generation and industries that use video or music.

It’s just a really narrow look at probably the worst use case for AI. Useless.

> I HOPE people aren’t spending money on AI writing emails.

They very definitely are. I'm seeing more and more front-line responses to customer support emails that are very clearly written by AI.

Here's an example from this week:

  Thank you for reaching out and providing all of those tracking and
  return details right away. I am currently checking in with our
  Returns Team to track down your package and see why there has been a
  delay in processing it since its delivery.

  As soon as I receive an update from our warehouse specialists, I
  will reach back out to you directly via this email thread with the
  status of your credit. Thank you so much for your patience in the
  meantime!

But why if that costing anyone money? You can generate like 50 of that kind of thing using various chatbots throughout your day.

You don’t need a pro account for emails.

Not that I've ever looked into it but I would have thought that the "free" chatbot AI services would prohibit commercial/business use without a paid plan. Maybe not?

No not at all

> I HOPE people aren’t spending money on AI writing emails. That’s definitely not worth it.

The engineers use the coding AI for everything they don't like to do that distracts them from doing what they like to do. So that's certainly a lot of writing emails, status updates, meeting notes, journaling, Slack messages, and prettifying slide presentations and so so so many docs.

So if you're aghast that hundreds of dollars of AI tokens per engineer per month is being spent on writing emails, you should be even more offended that the more expensive human engineer time used to be wasted on those tasks.

I use AI to shorten my emails, I write as I talk, and I ramble a lot in my head.

But surely the free version of any of the chatbots is enough to achieve that, right? You don’t need a Pro subscription for that.

It's not the only use case for LLMs I have. I use local LLMs for some things.

This comment is odd given that it's high on opinion without ever refuting the actual data given in the article.

Like it or not, many business people spend hours a day writing emails. The article pointed out specific data showing AI could reduce these hours spent while raising quality.

I'll take the article's actual data and research over your data-free spouts of "That's definitely not worth it" and "Useless".

I’m not saying “It’s crazy to ever write emails with AI.”

I’m saying it shouldn’t hardly cost any money, emails are not token intensive and most people could likely use the free version of chatbots just fine.

Except that many companies are pushing "Use AI for everything" -- seriously. Because the CEO was told by KPMG or whoever that "AI is the future". And so yeah, they're pushing it regardless of whether it makes actual sense to use it or not.

> Companies are spending the most on coding tools, not email writing.

They probably spend more on having LLMs just parse documents than writing emails than writing code. You maybe forget that most people are not programmers.

It doesn’t matter what “most people” are, coding agents use an obscene amount of tokens, literally millions per day. Every request to an agent needs to parse at least several different documents/files.

All LLMs use obscene tokens just programmers are more aware of it

That’s not true. The different use cases of LLMs objectively use different amounts of tokens. AI coding agents can parse half a dozen or more files on each request, plus use a ton for thinking.

Writing an email barely uses any in comparison.

> thinking

LLM doesn't think

> AI coding agents can parse half a dozen or more files on each request

yeh people never ever ask LLMs to write emails based on a number of docs and spreadsheets and other emails and Slack chats plus some websites and definitely won't iterate on it multiple times...

and parsing source code is peanuts compared to parsing a PDF

>LLM doesn't think

You don’t realize I’m referring to the process of generating a ton of text in an attempt to better address the request - a process that these LLM companies call “thinking”? Are you out of the loop or are you taking some kind of pointless stand here?

>yeh people never ever ask LLMs to write emails based on a number of docs and spreadsheets and other emails and Slack chats plus some websites and definitely won't iterate on it multiple times...

A worker might have one or two of those large requests a day. That stop doesn’t come close to agentic AI token usage. I’ve been applying for jobs daily, uploading both my resume and the job posting upwards of 6-7 times per day to these chatbots. I still never hit the free limit.

Agentic AI coding obviously uses far, far more. This really shouldn’t be surprising.

1. You seem to think that non programmers don't ask LLMs to program for them. Maybe that's why you're comfortable here thinking you're definitely will keep being in demand. Believe me they do ask an LLM to write for them various tools and bots and what LLM writes can in turn consume tokens like crazy be agentic or whatever the latest hype word is.

2. even without that uploading a resume is not "write emails based on a number of docs and spreadsheets and other emails and Slack chats plus some websites and iterate on it multiple times". one of those requests can easily be a month worth of your resume uploads at that rate.

>You seem to think that non programmers don't ask LLMs to program for them.

Because they largely do not. Are you actually under the impression that the average person is using chatbots to program? That’s wayyy off.

>Maybe that's why you're comfortable here thinking you're definitely will keep being in demand.

I’ve made no such claim, this a complete non-sequitur.

>Believe me they do ask an LLM to write for them various tools and bots and what LLM writes can in turn consume tokens like crazy be agentic or whatever the latest hype word is.

The average person does not do this. Come on, dude.

>even without that uploading a resume is not "write emails based on a number of docs and spreadsheets and other emails and Slack chats plus some websites and iterate on it multiple times".

It’s much closer to the average usage. All of that is still far less than agentic coding.

> using chatbots to program

They don't use chatbots to program. They use chatbots to get shit some. Sometimes that involves a program.

You know how a regular person can simply prompt a chatbot to generate a chatbot for them that would automate dating (incl messaging) and they get the full thing with handholding about what to press to get it running? Yeah. (If your model can't do it yet then try %someothermodel% or wait for a couple months.)

Your assumption that no one surely ever uses this for actual work purposes is funny. imagine you're preparing some slides and you need to process some data into it and the thing generates a script to do it. now imagine that generated script also calls some chatbot api and so on;)

3% of measurable gains is massive in itself isn't it? Taken as a percentage of the total volume of workforce and the world economy?

If we can gain 3% across the board gains on AI based tasks without subsidized expenses, that would be a great win!?

But is it a 3% gain if we include the costs? Or would it actually be cheaper to just hire humans with the money going to tokens?

And guess what humans even learn, get better and might bring more value in the future.

Citing a paper from 2023 and calling it frontier, nah

That study is a weird kind of bullshit which pretends to not understand employer-employee relationship.

Calling bullshit on this.

The size of projects I've done in the past year compared to before is mind boggling. There is NO way I could have reached this level of productivity without a properly trained and properly prompted LLM.

Like what

Their tool for sandboxing AI agents, probably.

You will never get an answer.

Can you say what exactly you've accomplished with this productivity? Can you provide revenue numbers and how profit increased or does your definition of productivity not include this?

Hell no. I want you to doubt. More for me.

I notice the study is "in the past" when models weren't as good and also it appears to focus on office workers writing text.

Ya, using AI to write your emails isn't that huge of a time savings. But coding is different matter.

> But in the wild the gains shrink to about 3% of your hours

not my experience at all

In research that studies a large enough population there will always be outliers. Maybe you are one of them. Your counterexample does not prove them wrong.

> AI genuinely speeds up the right tasks: writing, support, structured drafts.

Fascinating prose, implying that it’s possible that AI can inauthentically speed up the right tasks, which makes no sense at all. Good job Claude!

Just embarrassing to believe a study like this right now.

plot twist: the vast majority of "studies" are like this, complete garbage, but called "science" because they put a table with p-values in

There are a large amount of anti-AI articles being posted on the HN front page. It is not natural. There is a clear conspiracy, but why, and who is orchestrating it?