Because their core product is not a long-term sustainable business strategy. Local hardware and models will continue to improve to the point of not needing the hosted solutions. And if you do need a hosted solution, remember that the big cloud providers already offer these solutions, so signing up for OpenAI/Anthropic _and_ AWS/GCP/Azure is not a sound business decision compared to just signing up with 1 of them that offers your cloud infra + GenAI infra. (Which is why the long-term benefits for cloud companies will probably be for the likes of AWS and not the likes of OpenAI).
They'll continue to burn money for marginal model improvements in the next few years all the while having no moat _and_ having Open-Weight / Local models eat their lunch.
The only way for them to stay relevant as a company is to expand beyond simply providing the models.
I'm old enough to remember the arrival of RDBMS, once IBM primed the space with DB2.
There was a pitched battle over features like row-level locking as competitors like Sybase, Ingress and Oracle scrapped it out. New features arrived on a monthly cadence, with immense engineering effort behind them. The winners (Oracle mostly) won a great moat which led to them to where they are today.
The fact that so many AI companies can produce amazing coding tools so quickly shows there is no moat, supporting your theory.
That is impossible to fund. At some point someone will decide to stop throwing money on the firepit that's the current business model and then hardware prices crash back to earth as 60-70% of the global demand disappears overnight.
The hardware miniaturization gains have finally dried up, though.
I am pretty certain that the current state of the art silicon feature size won't shrink again for at least another decade or two.
It normally takes about a decade to mature a tech which can create a smaller feature size into something commercially viable for mass production scale, and no further improvements have been in the pipeline for that long now.
So it's like the "next piece" indicator while playing Tetris is just blank.
It is more of a regulatory capture byproduct to prop up an artificial token driven Ponzi scheme.
There is a serious alternative to NVIDIA "AI" hardware dropping out of China in February 2027. There is no moat, but a whole lot of unpaid debts in the near future.
I honestly have no expert knowledge about this stuff. What I say is based only on intuition.
- China is heavily, heavily incentivised to enhance their own chip making
- Looking at the rate Chinas has expanded into just about every single other
space, and from quantity to quality, I just think it is impossible that they
don't compete on equal grounds pretty soon.
- I don't buy the insurmountable moat of TSMC
Reminiscent of tank warfare in WW2. The soviet T-34 was not remarkable in any particular way, but the sheer volumes it was produced in made it a very serious enemy to German tanks. As Stalin said "quantity has a quality all of its own".
The companies who control the compute resources will ~always control the greatest "amount" of intelligence. They can lease that intelligence out, or they can use it themselves. Currently the "total amount of intelligence" or perhaps "total amount of ability-to-do-stuff" is split between humans and machines at a ratio that means it still makes sense to lease the machine intelligence to the human intelligence - plus there are things that humans are still better at. In maybe 2 more years that will stop being true, due to the availability of more physical compute resources, and far greater model intelligence per unit compute. At that point, the point at which the substantial majority of ability-to-do-stuff is controlled by machine intelligence, then the entities who control all the compute will control all the ability-to-do-stuff, i.e. "the economy."
So I agree that the core product is not long-term sustainable as a product but this is because the whole world will look so different in the near future that the framing of intelligence as a "product" breaks down.
Open-Weight models, of course, are fine and useful, but if you have one million times less compute than your competitor (the lab), then you're not really playing the same game. You can only tackle the problems that they have decided they're not interested in.
I don't know if the gap will close or rather widen with more compute coming online.
Being half a year to one year behind could be meaningful, not to mention that competitors may not have the necessary compute to train and serve models of a certain size.
This could be a significant advantage for OpenAI and Anthropic, and if they make breakthroughs in robotics or science, that is worth far more than mediocre coding assistants.
They do partner externally. This work is fundamental discovery science, rather than industrial research.
the folks who run anthropic grew up reading scifi with crazy awesome biotech. However, when they look at biotech today, it's just depressing. It's incredibly slow, it takes decadfes to prove out new technologies, and they figure with this new tool, they can just point it at problems and have it emit discoveries. If they show a few high-impact discoveries, that makes a case for them to move biotech forward much faster than its current progress.
Also, anthropic has so much capitalization right now that it's simply easiest to invest it in a wide portfolio that includes both internal and external research.
Who, you may ask, would take that money? People like business influencer Megan Lieu, who chose not to disclose just how much she'd made from her AI deals, but says her biggest sponsorship to date has been with Anthropic (makers of Claude), as well as that her biggest sponsored contracts (for any client) are normally around the $30,000 mark.
That isn’t actually the point, the point is that the AI companies should be made to see that doing things for society is the only way they get any kudos.
If they want to compete to be seen as the good guy, by all means let them. But it means actually having to be the good guy, in at least some respects.
I was actually thinking the other day that it makes perfect sense for AI companies to develop a professional services oriented software development arm. Imagine that you want to develop a training pipeline for "tasteful" programming: you might make a reward metric for that does some obvious stuff (nothing that anyone could easily agree is a bug like a crash, good performance, perhaps minimize LoC), but you really want to also want to also track "bugs" where the feature was discovered to be missing some unspecified nuance that was only discovered through product use, or train on ability to keep a small codebase while also keeping diffs small (essentially, "maintainability") as real new requirements come in.
So then you want a training set full of real product requirements and product evolution, which is something you could get if you offered custom software development, with a lot more control than you'd get trying to do the same by scraping random FOSS projects on github.
Other industries are perhaps similar. If you offer a service directly, you have much more ability to build collection of training data into the process. Want to make the best law bot? Buy a law firm, offer legal services, and integrate extremely deeply into their workflows. If their models turn out to be as good as they hype up, they should be able to scale to be a major player in any endeavor they move into with a relatively small number of staff and develop a strong feedback loop (not that that would be good for the rest of us).
If your core service is getting more expensive to provide and competitors are busy eating your margins, why let someone else taste your secret sauce and only get paid for the tokens, when you can keep the good stuff (bio capability) for yourself, and net both the profit and the fame?
I'm confused of why this is a question. First of all everyone is doing something because it benefits them. You and I included. Second of all as long as it's a real discovery, it will be beneficial to us all eventually (after benefiting Anthropic for sure).
Perhaps you're not on HN long enough, but there have been many posts where someone bemoaned the lack of basic science research by corporations, that IBM and Microsoft were the only a few remaining companies with any science research. Guess what? they do it for their own benefits as well.
Because as I see it, there are a lot of already established labs that could take research like this a lot further with the help of AI instead of just throwing more agents at the problem.
That’s my confusion around this topic. Does the strategy change when you can throw a bonkers amount of compute at the problem with fewer guardrails?
Because you're not understanding the goal. The goal isn't to assist humans in making the discovery. The goal is to develop a system that can autonomously make the discoveries, as this is way more scalable.
A chatbot for cancer researchers to talk to is worth single-digit billions at most. Anthropic is already valued at over a trillion dollars, on the premise that they can replace the majority of jobs in most knowledge industries. All the announcements about hacking / math problems / biological science are meant to create the impression that that strategy works and is repeatable across industries.
Cancer research is a lot harder for LLMs than math millennium problems though, because there is no fast feedback loop to iterate on. Even if you have a really good idea based on a solid theoretical insight, doing the experiments using in-vitro/mice/monkeys/humans can take years or even decades. I have no doubt that AI will help find new avenues that boost certain parts of research in these fields, but I don't see a potential for a drastic change until we at the very least give LLMs a direct way to interact with lab equipment and train them using RL on it.
Yes, but initial discovery of molecules and novel mechanisms is massive. That was the last generational change in modern drug research was the movement to high throughput screening, going from the ability to screen 10's of molecules to hundreds of thousands to find 'hits'. Better and more focused models, especially ones trained internally at big pharma companies will accelerate that portion of the pipeline, or increase the hit rate of successful compounds. Several companies are already taking this approach like Novo has been. There are other more early stage companies like Recursion and others that are doing the same thing. They are more tech companies than traditional wet lab companies.
Sorry, yes I agree 100%. I don't agree with their narrative, I was just explaining it. I think it's fraudulent and based on science fiction and will lead to a significant economic crisis.
Our lab, located in the Bay Area, looks like a typical molecular biology lab. We do research that involves only the lower-levels of the biosafety risk level (BSL-1 and BSL-2) and we do not handle pathogens that can infect humans. All of the lab work is performed by human scientists. Although we’ve experimented with using AI to accelerate lab work with initiatives like the Model Hardware Standard, this approach is less conducive to the sort of ad hoc workflows that are involved in our molecular biology research.
Automated labs are far more challenging to build and run than most people appreciate. If I wanted to make progress quickly in discovery science, I would find good lab techs before building automated labs.
> I’m confused why AI companies are using agents in-house for this type of research instead of partnering externally.
As an outsider, here is how I explain that behavior:
1. Truly risky models are very useful.
2. Truly risky models should not be released, according to AI safety standards. I think Antrhopic genuinely believes in AI safety. (see: standing up against automated kill chains, no matter the impacts to the company)
3. Truly risky models face regulatory pressures, if released to the public.
This all leads to "let's just do this in-house." I believe that might end up being the answer to every application of AI eventually. It seems unavoidable, and very depressing.
Lands as an active threat. Maybe they're serious about this research or not, but for sure medical companies doing this sort of research will consider upping their AI budget and connecting their labs, etc. to avoid "falling behind".
So, the AI labs benefit either from achieving something they could market or from the peer-pressure imposed to companies in the sectors they get their nose in.
I think the "everything company" vision has become apparent for a while now. Doesn't even have to be sinister - I think Anthropic simply believes on one else can be trusted with this power. Another point of leverage they have is that they can keep their internal models for themselves.
I run into this all the time - we have such powerful functionality available to our users, and further we provide the elements that undergird all of it, so it’s totally possible for clients to take the services they buy from us and reconfigure them to make their own tools, better even than the ones we have built, purpose-built for their workflows…
And 9/10 clients will just click on the one thing they know and recognize and are familiar with and comfortable with… and then stop thinking about it.
It’s crazy how much of our job is not only building our product, but interrogating our clients over what they need, so we can demonstrate how our tools solve their problem. The users simply are not interested in figuring it out for themselves.
This is my speculation as well. For the time being, knowing how to use Claude extremely effectively probably beats out industry insider status. And Anthropic can attract whatever expertise it needs to build scrappy research teams in house. I'm guessing this kind of work doesn't need 100+ people, maybe just a dozen highly specialized people.
Given the prestige of the AI labs, the recent explosion of math proofs, the literal millions they can throw around, it seems very likely they can attract then fund small research projects across a broad range of science. And like startup math, it only takes one or two ground breaking results from a hundred attempts to pay back in the PR/hype.
When I took web design courses they basically told us you effectively have to make two websites - one for IE6, and one for all the other browsers. It came down to a bunch of crazy hacks and conditionals you had to bake in.
I remember when in a past life we were allowed to bump the minimum supported IE version of our SPA framework from 6 to 8. It was a joyful day – but we soon realized that it didn’t really buy us that much because 8 was almost as quirky…
I wonder how many trillions in GDP we missed out on due to them building a monopoly in the browser space and then abandoning it. You can't imagine how happy I was in 2005 when Microsoft announced they were resuming development on IE. I was already a fan of Firefox by that point, but a lot of web dev was such a drag because you couldn't just use modern browser features--you had to accommodate all the users still browsing with that dinosaur.
It's not just that it had quirks, it's that your feedback loop was longer in IE compared to Firefox with Firebug and Chrome. IE 7 did not, to my recollection, have a way to open developer tools and tweak CSS values until it looked right, or even to see what the effective CSS on an element was. So you had to save and refresh. Sure, that's not a super long cycle time compared to, say, compiling a binary and flashing to a ROM, but it's longer than the other browsers I was working with at the time.
it was frustrating but i preferred it to dealing with stuff like node, tooling, mobile and tablet views, that stuff really burned me out in a way that IE 6 never did.
You've always been allowed to transcribe (ie, write) out a conversation. Recording is a different thing. A person can always claim, easily and believably, that the transcription is made up. It's just text.
From a legal perspective you can't record something in many us states. Using the microphone, keeping the bits in memory while you turn it into text, probably doesn't meet any reasonable definition of "record".
Maybe this will be true legally (we wont really know until it is tested in court), but I think there's a huge difference in practice.
Siri Recap is an electronic record of a conversation that includes timestamps. Imagine for example that during a scheduled meeting with my boss, he threatens to fire me for not doing something illegal. I now have strong evidence that this was said (even if it cant be officially attributed to him) during the time I was in a meeting with him. Not saying this will hold up in court, but it certainly is more damning that having written notes (not to mention the fact that my boss probably wouldn't say this when I am taking written notes).
Making that feature illegal may backfire spectacularly across the whole IT sector. What Apple does is it inputs voice data into an AI program and that outputs some transformed output. If a court would rule that output in any way matches the input (to be classified as a "record") then the whole ethics principle of stealing other people's data and funneling it through AI to make it company's own, would be at risk.
I imagine no single spineless impotent modern court would risk a wrath of our new benevolent AI overlords calling them all criminals they really are.
There was an app that hit it’s peak during covid: Clubhouse.
Many powerful and interesting people were having powerful and presumably private conversations in rooms because they say people walk in and out and legally CH could not record without consent. But what they could do was “collect metadata”. Emotional and tonal analysis is metadata. So is time, date, location, username (each user had their own mic stream), email, etc. Transcription is also metadata.
Essentially the tech has left the laws in the dust as you could now reasonably and legally record enough metadata to accurately resynthesize a recording.
If it doesn’t record anything how does it make a transcript? I think some of these laws are strict enough such that no audio recording, even if it’s immediately discarded, is legal.
There are similar exceptions. For instance, the copy of licensed content that your computer makes in RAM in order to play it is explicitly exempt from being treated as a copyright violation.
isnt that how the modern telephone system works? digitize your voice, encode, transmit, decode, convert to analog audio. with buffering or keeping bits in memory all along the way?
and taken to it's extreme.... isn't the atmosphere secretly taking the vibrations in the air right outside out mouth and propagating them through a physical medium, making them available to anyone to capture?
It's a toggle. Some will automatically enable it and you have to turn it off. People who rapidly click through setup flows can miss it and leave it enabled.
Wow it’s actually that bad…
I was looking for https://gruhn.me/blog/2026-08-03/ but it appears people have co-opted the original phrase and out SEOed the original article… only to make a less concise, AI generated copy of it mimicking nohello…. https://dontbeameatproxy.com
I had this debate with my coworker who prefers anthropic models to open ai ones. I ended up settling into the idea that gpt 5.6 is better used as a tool and opus 5 is a companion. GPT 5.6 takes you literally whereas opus 5 tends to take more liberties to try to get to the “spirit” of what you want. It comes down to preference, and I don’t want a companion.
That’s how I see it too. Claude is more “fun” to use, like a coworker I have to talk to now and then to steer it, while gpt-5.6 is a task machine: I give it a task and it is very consistent, reliable and predictable in its execution. I don’t have to interrupt it, it gets the task done exactly how I wanted it, but it’s “boring” and feels more sterile
Sol is an absolute machine. I stopped doing parallel worktrees just because the cost of context switch outweighs the cost of waiting Sol to just finish the task it’s working on which is usually anywhere from 1-10mins.
I also like Codex CLI more than the Codex App bc it’s more scriptable and displays all the tool calls and reasoning whereas in the App it’s kind of folded away/obscured. This way as soon as I see a tool call fail (eg it tries to use jq assuming it’s available but it wasn’t so I take a note to set it up as it’s obviously useful for the agent to wrangle json).
I think its amazing what OpenAI have been able to squeeze out from a model like Sol thats much smaller in size than Fable.
I think this is such a great reframing. It makes so much sense; I need an AI that acts more as a HUD and gives me superpowers, not just a copilot that can tell me when I've misspelled a word.
I've switched to Codex a few months ago when Claude's weekly limits were getting pretty stiff, and I haven't looked back. Both GPT 5.5 and 5.6 are quite capable, especially compared to nerfed Opus 4.7 (haven't tried 5).
Also, the Codex guy regularly resets weekly limits for everyone, which is a nice bonus (I know it's a temporary gimmick to attract more users, but I might as well use it while it lasts.)
One week it feels better to work with Fable and Opus 5, the other I work more with GPT 5.6 Sol. Either takes its liberties, and neither communicates like a companion.
I hear you guys, but it sounds like we’re taking about the default settings or “personalities” baked into the models by their creators.
Either of them will act exactly the way you want if you explicitly tell them too. Add the instructions to your own system prompt. If you don’t want a companion, say so. If you want shorter answers in a different style, tell them. They will obey :)
> We retired the “Nerdy” personality in March after launching GPT‑5.4. In training, we removed the goblin-affine reward signal and filtered training data containing creature-words, making goblins less likely to over-appear or show up in inappropriate contexts. Unfortunately, GPT‑5.5 started training before we found the root cause of the goblins. When we began testing GPT‑5.5 in Codex, OpenAI employees immediately noticed the strange affinity for goblins, and we added a developer-prompt instruction (opens in a new window) to mitigate. Codex is, after all, quite nerdy.
Note that the permanent solution was not just adjusting the prompt, and in fact being perfectly aware of that option they decided on a different course of action. That means either you are wrong or they are wrong.
> making goblins less likely to over-appear or show up in inappropriate contexts
So inappropriate goblins are still likely, just less so…
Hey, remember when tech bugs were things like buffer overflows or cross-thread performance impacts? I miss the days when our war with system goblins was purely metaphorical.
I guess the improvement loop is tighter and they have more control over how discoveries can be used for marketing?
But, in my mind, it begins to feel like they are setting themselves up to be “everything” companies instead of focusing on their core product…
reply