Not sure WTF I read here. Just more vibe coded "products" and "blogs", as it seems.
This "padded room" architecture fails because isolating the host OS does nothing to protect the user's data; if the agent has permission to read your files and access the internet, an injection will simply use the agent’s legitimate tools to exfiltrate your private information. Furthermore, making core memory files immutable and requiring manual confirmation for every action effectively lobotomizes the AI, trading its primary value—autonomy—for a false sense of security that users will eventually bypass due to click fatigue.
While that might take it a little too far, Lex surely is a dangerous individual. On various occasions did he sympathize with the war and terror that Russia is doing in Ukraine. I do not click on any of his content because I will not support these (and a few other questionable, to say the least) views of his. Also his image of an MIT researcher is hilarious.
> On various occasions did he sympathize with the war and terror that Russia is doing in Ukraine.
I'm not a devotee of his but I've listened to a few of his podcasts when I like the guest. I have an idea of how someone would come away with your impression given lex's interview style but I'd be pretty surprised if anything he said would, to me, fit your impression.
That said, I'd like an example if you have something specific to point to that might change my mind or if it's just a general takeaway you've gotten from a corpus of interviews on the topic (which would be totally valid but wouldn't change my mind).
> That said, I'd like an example if you have something specific to point to that might change my mind
This guy wanted Putin on his podcast to hear his side of the story (let that sink in) and spoke Russian to Zelensky. Willingly wanting to provide a platform for a mass murderer who is best known for large-scale social media propaganda.
This is not an "impression" of his "interview style". This guy implicitly supports terrorist acts.
> This guy wanted Putin on his podcast to hear his side of the story (let that sink in)
Many people have interviewed serial killers and not supported serial killers.
I would very much like to know Putin's actual motivations which would unlikely be spoken but his stated motives would also be enlightening.
I'm sure he'd go on with the standard "Nazis in Ukraine" line but in a 2-3 hour interview, I might get some new insights I don't get from 3 sentence sound bites.
We know so much about Hitler from his own writings and speeches. It seems to me that your philosophy on "platforming" Putin would also apply to making the words of Hitler available to the public.
Is there someone you think _could_ interview Putin responsibly?
> spoke Russian to Zelensky
I don't see the significance of that. They both speak Russian and English fluently. I don't know if Friedman speaks Ukranian but I'm not understanding what the implication is here. Surely the interview was in English since the podcast is?
> This is not an "impression" of his "interview style". This guy implicitly supports terrorist acts.
Implicitly being the key word here and is certainly subjective. If the body of evidence you're presenting is "would interview Putin" and "spoke Russian to Zelensky", I don't find that convincing.
> Is there someone you think _could_ interview Putin responsibly?
No, and no one should, see next answer.
> Implicitly being the key word here and is certainly subjective. If the body of evidence you're presenting is "would interview Putin" and "spoke Russian to Zelensky", I don't find that convincing.
"Would interview Putin" implies "is willing to provide a huge international platform for a terrorist and still-active mass murderer who is best known for his effective propaganda of peoples' minds". If you do not find that convincing, you are not alone at all. This has been the objective of Russia all along.
Pretty sure he’s a complete fraud too. He associates himself with MIT despite only having had a short stint teaching non-credit classes. One of his papers was apparently so flawed it’s been wiped from existence. Plenty of info online if you want to go down the rabbit hole.
So how many more LLM-generated "products" do we need to see here? This is inconsistent, with contradictory information, partly blatantly wrong and overall horrible from start to finish.
You might as well ask an accountant why he/she uses a calculator. I really don't think you're asking the right questions here. These questions will lead to obvious answers that you don't need to interview people for.
Consulting has weak margins compared to SaaS and scales poorly. Providing the interface for companies to spin up their own consultants (=Agents like Claude Code) is a superior business model in every dimension.
But those margins are for traditional businesses with human workers, if these claims of 100x productivity increase are real Anthropic should very easily be able to outcompete Accenture no?
Consulting - especially the more strategy type consulting - is often not about “we don’t know how to do something”, it’s more of “there is so much resistance to change organizationally that not even CxOs/directors can push it through”.
Besides selling consulting services involves a lot of relationship building and knowing the business vertical.
Only on HN will people still doubt what is happening right in front of their eyes. I understand that putting things into perspective is important, still, the type of downplaying we can see in the comments here is not only funny but also has a dangerous dimension to it. Ironically, these are the exact same people who will claim "we should have prepared better!" once the effects become more and more visible. Dear super engineers, while I feel sorry that your job and passion become a commodity right in front of you, please stay out the way.
Can you point me to a human written program an LLM cannot write? And no, just answering with a massively large codebase does not count because this issue is temporary.
> Can you point me to a human written program an LLM cannot write?
Sure:
"The resulting compiler has nearly reached the limits of Opus’s abilities. I tried (hard!) to fix several of the above limitations but wasn’t fully successful. New features and bugfixes frequently broke existing functionality.
As one particularly challenging example, Opus was unable to implement a 16-bit x86 code generator needed to boot into 16-bit real mode. While the compiler can output correct 16-bit x86 via the 66/67 opcode prefixes, the resulting compiled output is over 60kb, far exceeding the 32k code limit enforced by Linux. Instead, Claude simply cheats here and calls out to GCC for this phase (This is only the case for x86. For ARM or RISC-V, Claude’s compiler can compile completely by itself.)"[1]
I wish I could agree with you, but as a game dev, shader author, and occasional asm hacker, I still think AIs have demonstrated being perfectly capable of copying "those effects". It's been trained on them, of course.
You're not gonna one-shot RD2, but neither will a human. You can one-shot particles and shader passes though.
Depends on what we categorize as a coding agent. Devin was released two years ago. Cursor was about the same, and it released agent mode around 1.5 years ago. Aider has been around even longer than that I think.
Why do you believe an LLM can't write these, just because they're 3D? If the assets are given (just as with a human game programmer, who has artists provide them the assets), then an LLM can write the code just the same.
What? People can easily get assets, thats not a even a problem in 2026. Roller coaster tycoon's assets were done by the programmer himself. If its so easy why haven't we seen actually complex pieces of software done in couple of weeks by LLM users?
Also try building any complex effects by prompting LLMs, you wont get any far, this is why all of the LLM coded websites look stupidly bland.
Not sure what you're confused about, I never said assets were hard to get, I just said that the LLM needs to be provided a folder of the assets for it to make use of them, it's not going to create them from scratch (at least not without great difficulty, because LLMs are capable of using and coding Three.js for example). I don't know the answer to your first question because I don't hang around in the 3D or game dev fields, I'm sure there are examples of vibe coded games however.
As to your second question, it is about prompting them correctly, for example [0]. Now I don't know about you but some of those sites especially after using the frontend skill look pretty good to me. If those look bland to you then I'm not really sure what you're expecting, keeping in mind that the example you showed with the graphics are not regular sites but more design oriented, and even still nothing stops LLMs from producing such sites.
you have shown me 0 examples, I showed actual examples to the given question. Your answers have just been "AI can also do this" but gave no actual proof.
The examples are in the video I linked, as I said, if you don't bother to watch it then I'm not sure what to tell you. As I said for games I don't know and won't presume to search up some random vibe coded game if I don't have personal experience with how LLMs handle games, but for web development, the sites I've made and seen made look pretty good.
Edit: I found examples [0] of games too with generated assets as well. These are all one shot so I imagine with more prompting you can get a decent game all without coding anything yourself.
For someone speaking as you knew everything, you appear to know very little. Every LLM completion is a "hallucination", some of them just happen to be factually correct.
>> instead of execs telling us “you should do X”, we figure out what we think will have the most impact to our customers and work on building those features and tools
What do you mean? The quoted text is the exact strategy I always use.
I don't want or need to be told top down what to do, it's better to think for myself and propose that upward. Execs appreciate it because it makes their jobs easier; users get the features they actually want; I get to work on what I think is important.
I think this is the most efficient approach. Decisions should be made at the lowest possible level of the org chart.
However, it has an important assumption: You are sufficiently aware of higher level things. If you have a decent communication culture in your company or if you are around long enough to know someone everywhere, it should be fine though.
In my experience (I make tools for the network and security guys): that's why you don't propose only one thing. We often have one new project every year, we propose multiple ways to go about it, the leadership ask us to explore 2-3 solutions, we come back with data and propose our preferred solution, the leadership say 'ok' (after a very technical two-hour meeting) and propose minor alterations (or sometimes they want to alter our database design to make it 'closer' to the user experience...)
This can still be okay - but you have to be correct in a way that the company values. This of course needs to be without doing something against the rest of the company - either legally or sabotaging some other product are both out. Values is most commonly money, but there are other things the company values at times..
More often than not, things don't turn out too well if engineers decide what to build without tight steering from customers and/or upper management. This is exactly what it sounds like here. Tech for the purpose of tech. I understand this is HN and we have a pro-engineering bias here, at the same time, engineers don't tend to be the greatest strategists.
Customers and management should always be part of the loop. This is reflected in the original quote and my comment.
I just think that having to be micromanaged from the top down is completely miserable, is worse for the customer, and is time consuming for execs. It’s not a way to live.
You as an engineer should be familiar with users’ needs. I got into this field because I love automating solutions that help users solve their problems. So of course I want to know what they’re doing, and have a good idea of what would improve their lives further.
This "padded room" architecture fails because isolating the host OS does nothing to protect the user's data; if the agent has permission to read your files and access the internet, an injection will simply use the agent’s legitimate tools to exfiltrate your private information. Furthermore, making core memory files immutable and requiring manual confirmation for every action effectively lobotomizes the AI, trading its primary value—autonomy—for a false sense of security that users will eventually bypass due to click fatigue.
reply