There are a lot of commenters responding to this that appear to be confused about the Constitution and government power in the US.
The federal government has the power to regulate the conduct of private actors and to regulate the activity between private actors. The Civil Rights Act is an obvious example.
Relevant to this discussion, obviously HIPAA regulates how certain private actors must treat the data of other private actors, aka patients.
It's probably confusing because the first ten amendments to the Constitution, known as the Bill of Rights, primarily restrain government power. These were enacted because people believed that the original articles granted the federal government too much power. Amendments alter the existing document. Just like the articles in the original document, an amendment can also authorize government powers.
The 13th amendment is a clear example of an amendment that authorizes government power and restrains the conduct of private citizens. It abolished slavery and gave Congress the power to enforce the prohibition.
An amendment can absolutely be written to establish a right to privacy against other private actors and authorize Congress to enact regulation to enforce that right.
And even if you prefer to view the constitution as primarily setting up restrictions on government power, I think most of us here would agree that the at-scale data collection and surveillance our government performs under the guise of national security is an incredible power that is new and novel enough to warrant a discussion on keeping it in check.
The federal government a) saves everything, b) is constantly trying to attack encryption, c) tries to insert backdoors in hardware, and d) otherwise gives itself the means to spy on citizens and aliens alike. Even though we fight and have won, it’s been pyrrhic. I’m tired of constantly having the battle every time a new generation forgets why privacy is important.
I can’t imagine ways in which an amendment asserting every citizen’s innate ownership over their digital footprint, data sovereignty, etc. could be a misuse of the constitution. “No government shall abridge access to secure and private communication technology. No government shall collect or persist a profile of any citizen except as required to serve justice. No government can compel decryption of private data. Etc…”
The Claudeish drove me back to OpenAI and Codex as well. I find the Claudeish can be improved but it’s impossible to fully mitigate. Opus is unusable for anything I’m going to read.
“I would have to disagree that it gets as grating as Claudeish though.”
It’s hard to imagine anything more grating than Claudeish. To quote Rainer Wolfcastle, "My eyes! The goggles do nothing!"
Maybe. There are legitimate use cases. For example, data on what insurance a healthcare provider accepts is notoriously unreliable. There are startups that are using voice AI agents to call providers and ask them "hey are you still accepting these insurance plans?"
Obviously it would be great if the was a, ahem, legal consequence for not keeping this data up to date and available. But you know, if a frog had wings right?
How is that a legit case? The providers will hang up. The bot will take a maybe and assume yes. Why on earth would you trust this task to an agent when the consequences of a "my mistake you are absolutely right to call that out" could be hundreds of thousands of dollars?
Did you know that there are many, many steps between "do you take this insurance" and then someone performing hundreds of thousands of dollars of procedures on you?
Did you know that just going to a hospital for an emergency without checking it was in network can run you a 100k bill even if they do nothing but MRI and monitoring performed for one night?
You’re going to hit your out-of-network out-of-pocket max well before that. Will still be a big bill, but if you’re dying you won’t have much of a choice.
You know, it's pretty funny that we currently live in a world where it's much safer to trust AI with medical diagnosis than it is to trust AI to find and schedule a doctor visit for a human medical diagnosis.
Not saying to trust AI medical diagnoses. But the difference in risk (chance of failure times cost of failure) is surprisingly large, and in the wrong direction.
> data on what insurance a healthcare provider accepts is notoriously unreliable
To fix the health care system is possible.
To use AI to circumvent it just creates an incentive to make it worse. You spend more on AI, insurance companies spend even more money in surveillance to know who you are and to make things more complicated. It is a escalation that just ends making you spend more money to stay the same.
For example, data on what insurance a healthcare provider accepts is notoriously unreliable. There are startups that are using voice AI agents to call providers and ask them "hey are you still accepting these insurance plans?"
Why are you assembling a list of what health plans different doctors take? That sounds like a spamhaus scraping the real world for leads.
Every health care provider in America will tell you the same thing: "Call your insurance company to verify coverage." Every insurance company in America will tell you, "Call us to verify coverage."
Nobody outside of the patient, the doctor, and the payor (insurance company) has a need to know those things. Your "AI agent" certainly does not.
Because I want to go to the doctor and I don't want to personally call n shithole private equity owned chain practices myself to find out if they take my insurance and are accepting new patients?
There are healthcare concierge services for this, my job offers one, which I use any time I need something like this.
Your insurance company has a list of the doctors who take your insurance. That's their job. They have people who do nothing but that every day for their work.
and still see when I call Dentist, they tell me check with health insurance and ask them add us to portal then I call Dentist again and then make an appointment.
This whole cycle needs to be automated and will be soon!!
Same here, I work for AssemblyAI and still encounter the same problem. When I call a dentist, they tell me to check with my health insurance and ask me to add us to the portal. Then I have to call the dentist again to make an appointment.
In that case the startups should be able to defend themselves against the social stigma - "sure, there's a stigma against doing this, but here's why we think it's worthwhile anyway in this case".
Clearly you've never had shitty insurance and a need for a doctor. I've spent multiple hours on the phone trying to find someone that takes my insurance. It's an actual problem.
I don't think the parent comment disagrees with it being a problem. I read it as more: the use of an agent doesn't resolve the ambiguity in the underlying problem but instead injects another variable that could further misrepresent whether the person has valid insurance coverage (e.g. hallucinating or misunderstanding the nature or extent of coverage).
Same. My first real job out of college was at IBM 2004 to 2005. I was so proud to be an IBMer. Hated it. Lots of good people, but the bureaucracy, the pace, the whole thing, not fun.
AI products are going to have to deal with accounting. An agent performs some work. It uses some models and perhaps some other metered services. Someone has to meter that activity. They have to attribute costs, apply the product's pricing rules, bill the customer, collect, reconcile with vendors, and maintain the ledger.
Stripe can use OpenRouter to build the financial and accounting infrastructure for every product that sells metered AI work.
I think the analogy is ADP. Payroll for all the work that's going to be done by AI agents.
A single policy location for multiple models with usage limits/rate limits, spend limits all configurable using a common syntax + deep integration with account levels and high visibility into subscription/payments related to meter usage to help understand whether or not your pricing model is underwater as costs change is exceptionally valuable. AFAIK each product consuming model providers without an openrouter style proxy has to do that work themselves, and if you are already using stripe for subs/payments, the direct integration is a huge value add for customers, especially if you can alert on "upcoming price changes from your provider changes your unit economics for XYZ packages/sub tiers".
Similar to how Stripe is a middleman for payments across (fragmented) banks, they want to be a middleman for (fragmented) AI models as tokens are the new currency.
As AI agents/harness/human are spenders of tokens, enabling them to derisk from being locked to a specific model provider & allowing to (re)route to any model at anytime for better leverage in a single API, similar to how they are doing for payments.
paying for data access, for "deterministic computing", for letting the AI use online services. Maybe even micro payments for using tested and verified skills/mcps.
Example: access to real-time stock trading data, access to weather information, letting it make stock trades, etc.
At first I was confused why Stripe bought OpenRouter, but I think this makes sense.
Not OP, but I could see the vision. Right now OpenRouter is developer-paid with no way to pass costs onto the end user. Stripe could help bill and meter at the user level for per-user billing. I would also think more SDKs/easier ways to add AI features (paid for via openrouter) into apps and websites.
Open router supports per-request usage metering, as well as per-key. Building user-level metering from this is trivial, and there are many implementations of this out there. Supporting per-user first-class is trivial considering Open router already has per-request and already supports per-request attribution. How important that is, is a big question, but definitely not something that requires Stripe to build.
Could there be use cases where you want to run inference for users in your app, but don't want to host your own backend, user authentication, and billing?
The user could log into his personal OpenRouter account, authorize your application, and optionally set a budget, all in something like a Stripe payment screen.
> The user could log into his personal OpenRouter account, authorize your application, and optionally set a budget, all in something like a Stripe payment screen.
Good point, just needs an user friendly flow to sign up and set the budget in a way that is more similar to Stripe checkout than regular account creation.
Perhaps stripe thinks metering is a lovely big market? A nice complement to their existing service. Plus a little AI buzz likely helps their valuation. OpenRouter could be a step rather than a goal.
As user agents become more common it's only natural that they will be used for taking the heavy lifting out of e-commerce purchases. There will be a big need for digital payments to verify and reconcile these purchases.
With LLMs that are plugged into digital payments we will essentially have buying agents in our pocket that can find us exactly what we want for the cheapest price and the quickest delivery.
I'm always surprised at how many people in the tech sector don't think that things can and will change. In the next couple of years? Sure. 5+? Who knows. Necessity is the mother of invention.
I'm always surprised at how many people in the tech sector assume overcoming fundamental LLM weaknesses is always “just a matter of time,” and that time will arrive quickly enough to be relevant. If you haven’t seen any of the videos of people trying to get frontier LLMs to count consistently, you might think we’re further from LLM accountant than you realize. It’s not a data problem— I’m pretty sure they’ve encountered numeric progressions between 1 ABs l and 100 before.
I'm not just talking about LLMs though. People thought the internet and email were just a fad that would never catch on. People thought that the idea of home computing was absurd. All because they only saw what was in front of them at the time.
Also, how many accountants do you know that are truly happy with their work? I'd like to think many of them would love to do something other than crunch numbers all day.
Ok, let’s cherry pick some more societal tech revolutions. People also thought Segways were not the revolution in transportation despite fans and industry proponents saying we’d need to redesign cities to accommodate them. People also thought blockchain ledgers were a niche, even if useful, technique despite fans and industry proponents saying they were going to change literally everything.(remember Web 3.0?) People thought NFTs were a fundamentally flawed solution in search of a problem despite fans and industry proponents asserting they’d revolutionize IP. People thought that the metaverse was kind of ridiculous despite fans and industry proponents saying we’d all live a significant portion of our lives there, and some were practically running victory laps at the beginning of the pandemic when everybody was forced online (but then still didn’t use the metaverse. Some thought immersive VR headsets were a niche technology with limited consumer appeal despite fans and industry proponents claiming they’d essentially be the default display for computing by now.
> People thought the internet and email were just a fad that would never catch on. People thought that the idea of home computing was absurd.
We're far removed from "we've sold all the computers the world can buy" (back from when computers were the size of a building).
This is a disingenuous argument. The web and email were basically instant hits and people realized it. Similar story for home computing once the computing power caught up.
Email was a hit well before the internet existed. The utility was immediately apparent. And while people had lots of concerns about the internet, like misinformation and consumer-level utility, it was never broadly considered to be as dubiously useful or ethically fraught as LLMs and diffusion models. You can definitely find articles criticizing the bubble-forming finances, and one or two written by dedicated serial naysayers, but it was not a common perspective. Anybody who lived through that era will remember a hell of a lot more pushback against mobile phone adoption than Internet adoption, and even that was culturally marginal.
> I'm always surprised at how many people in the tech sector assume overcoming fundamental LLM weaknesses is always “just a matter of time,” and that time will arrive quickly enough to be relevant.
Lots of techies are tech-optimists ("tech always improves quickly").
Lots of people also have dollar signs in their eyes (or related, such as increased visibility and scope).
I think those are more like two facets in the crowd rather than two distinct crowds. Almost all idealist tech industry workers in column A that believe this is the future of the tech business would also fit into column B. Column B folks believing this business isn’t a fantastically expensive mass delusion event (with regard to the financial impact predictions) would essentially fit into column a. Most that wouldn’t would probably be Wall Street types that told Leopold Aschenbrenner to get bent rather than handing him billions of dollars on blind faith.
Things can easily change for the worse, such as a stock market crash. Who knows? Will we get AGI or will we see more homeless? History says... AGI for sure!
It's very difficult for this sort of thing to change for the worse. It's basically impossible on the software side. If next year's model is worse then keep using this year's model. Hardware technology can potentially go backwards, but things have to get really bad for that to happen.
LLMs are inherently unreliable. We do already work with unreliable technologies for many things but there's this engineering principle called "use the right tool for the job".
And it's a lot more likely that LLMs can't be changed to become unreliable, it's just how they work. So we would need more basic research, that doesn't grow on trees and for which the timelines are basically open ended. Maybe tomorrow, maybe right after cold fusion hits mass adoption.
They’ve become a lot more reliable. It would be strange if that progress stopped now. It’s possible. My main point is that they won’t get worse, as the other commenter suggested.
This isn’t a theoretical or technical problem; in fact, it’s not about the models at all: it’s about the available services getting worse. When they preview a new model and people are benchmarking it, they afford them a shitload of compute and they work great. Then, they get worse. That extra compute is surely in the billions of dollars OpenAI allotted to marketing, and especially in higher-volume periods, people talk about Anthropic‘s responses being lower quality all the time. The finances don’t indicate these companies’ current MOs are sustainable. Since self-hosted models don’t stand a chance of competing with frontier model capabilities, there is a very real chance that the available services will get much worse, and where the rubber hits the road, that’s really all that matters in the foreseeable future.
Everyone seems to be asking why this a great strategic fit for Stripe.
Stripe can now provide tools to every product that sells metered AI usage and take a cut. This isn't about extracting a small percent on the tokens flowing from your coding agents to your model of choice. It's about all the products that are going to come to market and monetize metered usage.
Want to analyze your 2026 tax return? Use VisorAI's Tax Agent and pay only for what you use. Stripe provides all the accounting, payment processing, ships money to vendors, and takes a bit off the top. Trillions of dollars moving from the labor market to the token market? It's $10B per point in fees for every trillion.
Ha, no I just tried to think of an example and an accountant wearing a visor turned AI came to mind. Pardon my hallucinations. It happens to meat sacks too.
Of course, of course, it turns out that name is already taken. Though not anything accounting related. A million monkeys all armed with a million typewriters is quite something.
It seems that way but I don't think it is. That user hasn't posted any similar promotional-seeming messages in any other recent comments. I believe they were meaning to offer an example to illustrate the point they're making about how Stripe can generate revenue by owning intermediary APIs.
The submission is cool because it opens the comments even if you just happen upon some article that also happens to have a HN discussion. That said, when I read the headline I thought the same thing, “uh, right click, open in split view?” When that feature landed in Chrome my first use case was exactly this.
Even if you can get it approved you are adding surface area to your annual security audits, adding another vendor that needs to be disclosed on security assessments, spreading your data to yet another processor, and adding another invoice and budget discussion. Depending on your customer contracts you may need to notify them of a new vendor. This might trigger a new security review. Oh it’s just another model on Bedrock? Bliss.
The federal government has the power to regulate the conduct of private actors and to regulate the activity between private actors. The Civil Rights Act is an obvious example.
Relevant to this discussion, obviously HIPAA regulates how certain private actors must treat the data of other private actors, aka patients.
It's probably confusing because the first ten amendments to the Constitution, known as the Bill of Rights, primarily restrain government power. These were enacted because people believed that the original articles granted the federal government too much power. Amendments alter the existing document. Just like the articles in the original document, an amendment can also authorize government powers.
The 13th amendment is a clear example of an amendment that authorizes government power and restrains the conduct of private citizens. It abolished slavery and gave Congress the power to enforce the prohibition.
An amendment can absolutely be written to establish a right to privacy against other private actors and authorize Congress to enact regulation to enforce that right.
reply