Anecdotally, I’m finding that, at least in the Spark ecosystem, AI-generated ideas and code are far from optimal. Some of this comes from misinterpreting the (sometimes poor) documentation, and some of it comes from, probably, there not being as many open source examples as CRUD apps, which AI “influentists” (to borrow from TFA) appear to often be hyping up.
This matters a lot to us because the difference in performance of our workflows can be the difference in $10/day in costs and $1000/day in costs.
Just like TFA stresses, it’s the expertise in the team that pushes back against poor AI-generated ideas and code that is keeping our business within reach of cash flow positive. ~”Surely this isn’t the right way to do this?”
I never read the tweet as anything other than that an expert with deep knowledge of their domain was able to produce a PoC. Which I still find to be very exciting and worthy of being promoted. This article didn't really debunk much.
My anxiety about falling behind with AI plummeted after I realized many of these tweets are overblown in this way. I use AI every day, how is everyone getting more spectacular results than me? Turns out: they exaggerate.
Here are several real stories I dug into:
"My brick-and-mortar business wouldn't even exist without AI" --> meant they used Claude to help them search for lawyers in their local area and summarize permits they needed
"I'm now doing the work of 10 product managers" --> actually meant they create draft PRD's. Did not mention firing 10 PMs
"I launched an entire product line this weekend" --> meant they created a website with a sign up, and it shows them a single javascript page, no customers
"I wrote a novel while I made coffee this morning" --> used a ChatGPT agent to make a messy mediocre PDF
Idk man, all AI discussion feels like a waste of effort.
“yes it will”, “no it won’t” - nobody really knows, it's just a bunch of extremely opinionated people rehashing the same tired arguments across 800 comments per thread.
There’s no point in talking about it anymore, just wait to see how it all turns out.
There are two major reasons people don't show proof about the impact of agentic coding:
1) The prompts/pipelines portain to proprietary IP that may or may not be allowed to be shown publically.
2) The prompts/pipelines are boring and/or embarrassing and showing them will dispel the myth that agentic coding is this mysterious magical process and open the people up to dunking.
For example in the case of #2, I recently published the prompts I used to create a terminal MIDI mixer (https://github.com/minimaxir/miditui/blob/main/agent_notes/P...) in the interest of transparency, but those prompts correctly indicate that I barely had an idea how MIDI mixing works and in hindsight I was surprised I didn't get harrassed for it. Given the contentious climate, I'm uncertain how often I will be open-sourcing my prompts going forward.
Did you post them with commentary along the lines of "this is the second coming of $DEITY, AI will replace us all, click on this Claude referral link to sign up"?
No, don't think so.
However, 90% of "AI" articles either are full of bullshit about "AI" or are someone trying to pass as an "expert" in some domains with LLM generated bullshit.
Doesn't the existence of consumer products like ChatGPT indicate that LLMs aren't able to do human-level work? If OpenAI really had a digital workforce with the capabilities of ~100k programmers/scientists/writers/lawyers/doctors etc, wouldn't the most profitable move be to utilize those "workers" directly, rather that renting out their skills piecemeal?
This is a strage phenomenon where people get excited by the mere fact that someone else is excited by something which is not directly visible to the spectator. It works well in horror movies and as it seems with AI hype.
Its still not a Hype, its still crazy what is possible today and we still have no clear at all if this progress continues as it does or not with the implication, that if it continues, it has major implications.
My wife, who has no clue about coding at all, chatgpted a very basic android app only with guidance of chatgpt. She would never ever been able to do this in 5 hours or so without my guidance. I DID NOT HELP HER at all.
I'm 'vibecoding' stuff small stuff for sure, non critical things for sure but lets be honest, i'm transforming a handfull of sentences and requirements into real working code, today.
Gemini 3 and Claude Opus 4.5 def feel better than their prevous versions.
Do they still fail? Yeah for sure but thats not the point.
The industry continues to progress on every single aspect of this: Tooling like claude CLI, Gemini CLI, Intellij integration, etc., Context length, compute, inferencing time, quality, depth of thinking etc. there is no current plateau visible at all.
And its not just LLMs, its the whole ecosystem of Machine Learning stuff: Highhly efficient weather model from google, Alpha fold, AlphaZero, Roboticsmovement, Environment detection, Image segmentation, ...
And the power of claude for example, you will only get with learning how to use it. Like telling it your coding style, your expectations regarding tests etc. We often assume, that an LLM should just be the magic work collegue 10x programmer but its everything an dnothing. If you don't communicate well enough it is not helpful.
And LLMs are not just good in coding, its great in reformulating emails, analysing error messages, writing basic SVG files, explaining kubernetes cluster status, being a friend for some people (see character.ai), explaining research paper, finding research, summarizing text, the list is way to long.
Alone 2026 there will go so many new datacenters live which will add so much more compute again, that the research will continue to be faster and more efficient.
There is also no current bubble to burst, Google fights against Microsoft, Antrophic and co. while on a global level USA competets with China and the EU on this technology. The richest companies on the planet are investing in this tech and they did not do this with bitcoins because they understod that bitcoin is stupid. But AI is not stupid.
Or Machine learing is not stupid.
Do not underestimate the current status of AI tools we have, do not underestimate the speed, continues progress and potential exponential growth of this.
My timespan expecation for obvious advancments in AI is 5-15 years. Experts in this field predict already 2027/2030.
But to iterate over this: a few years ago no one would have had a good idea how we could transform basic text into complex code in such a robust way, which such diverse input (different language, missing specs, ...) . No one. Even 'just generating a website'.
I’ve taken to calling this (in my mind) the Age of the Sycophants. In politics, in corporate life, in technology and in social media, many people are building a public life around saying things that others want to hear, with demonstrably zero relationship to truth or even credibility.
Like everything in LLM land it's all about the prompt and agent pipeline. As others say below, these people are experts in their domain. Their prompts are essentially a form of codifying their own knowledge, as in Rakyll and Galen's examples, to achieve specific outcomes based on years and maybe even decades of work in the problem domain. It's no surprise their outputs when ingested by an LLM are useful, but it might not tell us much about the true native capability of a given AI system.
Great article. This needs to be framed. The whole trust me bro, and shock and awe of social medias is a non-stop assault these days. You can't open a wall without seeing those promoted up front and centre and without any proof.
If AI was so good today, why isn't there an explosion of successful products? All we see is these half baked "zomg so good bro!" examples that are technically impressive, but decisively incomplete or really, proof of concepts.
I'm not saying LLMs aren't useful, but they're currently completely misrepresented.
Hype sells clicks, not value. But, whatever floats the investors' boat...
Its a strange phenomenon. You want to call out the bs but then you are just giving them engagement and boost. You want to stay away but there is a sort of confluence where these guys tend to ride on each others' post and boosts those posts anyway. If you ask questions, very rarely they answer, and if they do, it takes one question to unearth that it was the prompt or the skill. Eg: huggingface people post about claude finetuning models. how? when they gave everything in a skill file, and claude knew what scripts to write. Tinker is trying the same strategy. (yes, its impressive that claude could finetune, but not as impressive as the original claim that made me pay attention to the post)
It does not matter if they get the details wrong, its just that it needs to be vague enough, and exciting enough. Infact vagueness and not sharing the code part signals they are doing something important or they are 'in the know' which they cannot share. The incentives are totally inverted.
If you don't get the results you don't get the results. If someone else can use this tool to get the results, they'll out-compete you. If they can't, then they've wasted time and you'll out-compete them. I see these influencer guys as idea-generators. It's super-cheap to test out some of these theories: e.g. how well Claude can do 3D modeling was an idea I wanted to test and I did and it's pretty good; I wanted to test Claude as a debugging aid and it's a huge help for me.
But I would never sit down to convince a person who is not a friend. If someone wanted me to do that, I'd expect to charge them for it. So the guys who are doing it for free are either peddling bullshit or they have some other unspecified objective and no one likes that.
The article nails the pattern but I think it's fundamnetally an incentives problem.
We're drowning in tweets, posts, news... (way more than anyone can reasonably consume). So what rises to the top? The most dramatic, attention-grabbing claims. "I built in 1 hour what took a team months" gets 10k retweets. "I used AI to speed up a well-scoped prototype after weeks of architectural thinking" gets...crickets
Social platforms are optimized for engagement, not accuracy. The clarification thread will always get a fraction of the reach of the original hype. And the people posting know this.
The frustrating part is there's no easy fix. Calling it out (like this article does) get almost no attention. And the nuanced followup never catches up with the viral tweet.
A recent favorite of mine is simonw who usually is unable to stop hyping LLMs suddenly forgetting they exist in order to rhetorically "win" an argument:
> If you're confident that you know how to securely configure and use Wireguard across multiple devices then great
Almost every aspect of public life on social media nowadays is guided by sensationalism. It's simply a numbers game, and the "number" is engagement. Why would you do anything that's not completely geared towards engagement?
Being respected inside big companies has little to do with engagement on social media. Most of the best engineers are working hands-down. Arguably shitposting on the internet may have a negative correlation with technical ability inside Google.
One of the times I think the draconian approach Apple has towards employee speaking as an associate of the firm without explicit authorization is the correct one.
49 comments
[ 5.6 ms ] story [ 83.3 ms ] threadThis matters a lot to us because the difference in performance of our workflows can be the difference in $10/day in costs and $1000/day in costs.
Just like TFA stresses, it’s the expertise in the team that pushes back against poor AI-generated ideas and code that is keeping our business within reach of cash flow positive. ~”Surely this isn’t the right way to do this?”
That said, I use Antigravity with great success for self hosted software. I should publish it.
Why haven't I?
* The software is pretty specific to my requirements.
* Antigravity did the vast amount of work, it feels unworthy?
* I don't really want a project, but that shouldn't really stop me pushing to a public repo.
* I'm a bit hesitant to "out" myself?
Nonetheless, even though I'm not the person, I'm surprised there isn't more evidence out there.
Here are several real stories I dug into:
"My brick-and-mortar business wouldn't even exist without AI" --> meant they used Claude to help them search for lawyers in their local area and summarize permits they needed
"I'm now doing the work of 10 product managers" --> actually meant they create draft PRD's. Did not mention firing 10 PMs
"I launched an entire product line this weekend" --> meant they created a website with a sign up, and it shows them a single javascript page, no customers
"I wrote a novel while I made coffee this morning" --> used a ChatGPT agent to make a messy mediocre PDF
“yes it will”, “no it won’t” - nobody really knows, it's just a bunch of extremely opinionated people rehashing the same tired arguments across 800 comments per thread.
There’s no point in talking about it anymore, just wait to see how it all turns out.
1) The prompts/pipelines portain to proprietary IP that may or may not be allowed to be shown publically.
2) The prompts/pipelines are boring and/or embarrassing and showing them will dispel the myth that agentic coding is this mysterious magical process and open the people up to dunking.
For example in the case of #2, I recently published the prompts I used to create a terminal MIDI mixer (https://github.com/minimaxir/miditui/blob/main/agent_notes/P...) in the interest of transparency, but those prompts correctly indicate that I barely had an idea how MIDI mixing works and in hindsight I was surprised I didn't get harrassed for it. Given the contentious climate, I'm uncertain how often I will be open-sourcing my prompts going forward.
No, don't think so.
However, 90% of "AI" articles either are full of bullshit about "AI" or are someone trying to pass as an "expert" in some domains with LLM generated bullshit.
Stuff like yours is rare.
My wife, who has no clue about coding at all, chatgpted a very basic android app only with guidance of chatgpt. She would never ever been able to do this in 5 hours or so without my guidance. I DID NOT HELP HER at all.
I'm 'vibecoding' stuff small stuff for sure, non critical things for sure but lets be honest, i'm transforming a handfull of sentences and requirements into real working code, today.
Gemini 3 and Claude Opus 4.5 def feel better than their prevous versions.
Do they still fail? Yeah for sure but thats not the point.
The industry continues to progress on every single aspect of this: Tooling like claude CLI, Gemini CLI, Intellij integration, etc., Context length, compute, inferencing time, quality, depth of thinking etc. there is no current plateau visible at all.
And its not just LLMs, its the whole ecosystem of Machine Learning stuff: Highhly efficient weather model from google, Alpha fold, AlphaZero, Roboticsmovement, Environment detection, Image segmentation, ...
And the power of claude for example, you will only get with learning how to use it. Like telling it your coding style, your expectations regarding tests etc. We often assume, that an LLM should just be the magic work collegue 10x programmer but its everything an dnothing. If you don't communicate well enough it is not helpful.
And LLMs are not just good in coding, its great in reformulating emails, analysing error messages, writing basic SVG files, explaining kubernetes cluster status, being a friend for some people (see character.ai), explaining research paper, finding research, summarizing text, the list is way to long.
Alone 2026 there will go so many new datacenters live which will add so much more compute again, that the research will continue to be faster and more efficient.
There is also no current bubble to burst, Google fights against Microsoft, Antrophic and co. while on a global level USA competets with China and the EU on this technology. The richest companies on the planet are investing in this tech and they did not do this with bitcoins because they understod that bitcoin is stupid. But AI is not stupid.
Or Machine learing is not stupid.
Do not underestimate the current status of AI tools we have, do not underestimate the speed, continues progress and potential exponential growth of this.
My timespan expecation for obvious advancments in AI is 5-15 years. Experts in this field predict already 2027/2030.
But to iterate over this: a few years ago no one would have had a good idea how we could transform basic text into complex code in such a robust way, which such diverse input (different language, missing specs, ...) . No one. Even 'just generating a website'.
If AI was so good today, why isn't there an explosion of successful products? All we see is these half baked "zomg so good bro!" examples that are technically impressive, but decisively incomplete or really, proof of concepts.
I'm not saying LLMs aren't useful, but they're currently completely misrepresented.
Hype sells clicks, not value. But, whatever floats the investors' boat...
Sitting 2 hours with an Ai agent developing end to end products does.
It does not matter if they get the details wrong, its just that it needs to be vague enough, and exciting enough. Infact vagueness and not sharing the code part signals they are doing something important or they are 'in the know' which they cannot share. The incentives are totally inverted.
But I would never sit down to convince a person who is not a friend. If someone wanted me to do that, I'd expect to charge them for it. So the guys who are doing it for free are either peddling bullshit or they have some other unspecified objective and no one likes that.
We're drowning in tweets, posts, news... (way more than anyone can reasonably consume). So what rises to the top? The most dramatic, attention-grabbing claims. "I built in 1 hour what took a team months" gets 10k retweets. "I used AI to speed up a well-scoped prototype after weeks of architectural thinking" gets...crickets
Social platforms are optimized for engagement, not accuracy. The clarification thread will always get a fraction of the reach of the original hype. And the people posting know this.
The frustrating part is there's no easy fix. Calling it out (like this article does) get almost no attention. And the nuanced followup never catches up with the viral tweet.
> If you're confident that you know how to securely configure and use Wireguard across multiple devices then great
https://news.ycombinator.com/item?id=46581183
What happened to your overconfidence in LLMs ability to help people without previous experience do something they were unable to before?
This is truly impressive and not only hype.
Things have been impressive at least since April 2025.
One of the times I think the draconian approach Apple has towards employee speaking as an associate of the firm without explicit authorization is the correct one.