From a user’s perspective 4.7 is a downgrade compared to 4.6 . It’s intended to give Anthropic more control about their compute resources and profitability:
We have developed a simple API that can produce tokens and events in various formats like jsonl, sse, even csv. Cancelation can happen when the socket is closed when streaming - fully automated - or when you push an event from another endpoint so that we stop the stream midway.
Background task are also subscriable and cancelable.
100% - the argument of the article is that building any feature beyond chat-based-demos on HTTP SSE streaming is super complex. But a lot of folks still want to do it, because that's what their tech stack is. I think it's still a valuable thing to be talking about how you might do that.
Coding agents can write ASM. But if you mean writing the actual byte-code that will require a very different approach at a very different level of abstraction that LLMs are not designed to do. Keep in mind that all LLMs are trained first on text and then fine-tuned on code.
> Models favor monolithic, single-file implementations that diverge sharply from human-written code.
Well, all of our code is monolithic with some files close 20K lines of code and we do use coding agents - not for the original code but as of late. I've always had that hunch that splitting everything into tiny files does not improve AI coding agent performance although it feels counterintuitive due to model context constraints.
To me the important parts of a program should be clustered together so the implementation is obvious. Scattering the implementation in various files all over the source tree does not help much building the mental model.
That also closely match how software used to be written in the past too.
Kinda surprising to me, since i had some trouble with Cursor & Co. once the file went over ~800 lines. It repeatedly failed to edit it, until i split it up into multiple logical components. As it should have been from the beginning...
Though, it was some time ago, so things might have improved?
VSCode basically any model can edit the 20K file without any issues. The coding harness does not read the entire file at once though. It reads chunks of it so the size does not really matter. What matters is how close are the things the agent needs to make the edit.
Yeah, that was my experience with Grok, whenever I gave it a file with over 400 lines it would just fail to comprehend it or be too lazy to write too much at a time. Splitting stuff up into separate files helped.
> Models favor monolithic, single-file implementations that diverge sharply from human-written code.
This isn't the case if models are prompted to actually plan the file architecture beforehand, it's only the case if they're given a dumb monolithic "code this thing" prompt.
> Scattering the implementation in various files all over the source tree does not help much building the mental model.
Yeah, that happens where I work and I hate it. A combination of lint rules and AI reviewer prompts complain about long files and long functions. This means something that could be a 300 line self contained function that could be read linearly, gets split up into 6 functions across 6 files.
It's the illusion of "clean code". If you're casually skimming the code, you feel good. But as soon as you go beyond the surface level it becomes annoying.
The so called AI job loses are not due to AI. I don't think there is anyone out there to argue otherwise.
In a year but probably sooner, when software systems start collapsing, and they will, hiring in tech sector will skyrocket. In fact, I don't believe the world have enough developers to backfill for the AI deficiencies.
To me the math is obvious. Assuming humans touch a 1% of all software systems created, something we know it is simply never going to be true given the current state and upcoming regulations, the 47 million developers world-wide (and that includes all kinds of developers) are simply not enough.
However, although jobs will be back and it will be better payed, programming will "suck" even more and I don't think it will be for everyone. If you are not the kind of person that enjoys reversing a piece of tangled mess it might not be for you.
If AI is everything and AI is software then everything is software and everyone would like to have a piece of that software.
Vanity titles never make much sense, and now even more people can call themselves “engineers”. I was always at a loss why many weren’t calling themselves “web engineers”. Hey Mom, I used Claude Code today at work so I’m an Agentic Engineer!
Imagine an agent dropping a directory with 1m images in it. just figuring out what happened and what got dropped, restoring it one by one, etc. - doable, but ergonomics are a bit lacking.
The reason this blog post does not come with any concrete examples how to use this enablement for useful and constructive things tells you something very important - it is a toy and they do not know who and how they will use it.
It is cool feature but to what end? Buying a domain is not something you have to do daily to require any kind of automation.
I am also not sure who Stripe Atlas for. I am genuinely confused. It is definitely not something a developer will use.
I understand that you can bootstrap a number of systems but that is like half-hour of work and arguably it is probably a good idea to do it manually to make sure you have strong foundations.
I've have personally never seen a good example where a cross vendor account provisioning actually working. For example, Fly.io used to provision Sentry accounts automatically which you could not access in any other way but through Fly.io. I mean the Sentry account was effectively locked to a project that you cannot transfer - hijacking the actual global alias as well. Vercel did something similar with PostgreSQL via Neon and Redis via Upstash resulting in painful migration processes.
I can imagine ending in some kind of deadlock between services due to security hence why the 30 minutes initial setup is kind of time well spent to avoid future issues.
Perfect for spammers, scammers and domain squatters, who can now automate their activities even more.
Can’t think of any other uses for this given the current state of LLM ‘agents’, though I can’t wait for the next report of something like ‘openclaw registered 1000 domains for me without asking and now cloudflare won’t refund me’.
LLM generation in general provides the most use to scammers and the like. Generate emails which people won't read, generate articles which are just honeypots or rip-offs, generate images to said articles, generate more and more spam.
Every legit use case for LLM practically requires that human would verify the result manually, at least briefly. But spammers can enjoy skipping that step, since content was never a main priority in the first place.
So, so many pro-AI/AI boosters just... Ignore this rather inconvenient fact in my experience. They will hype up how epic agentic coding is, or agentic <whatever here> is, all day long, but they will never tell you that LLMs are really benefiting scammers and criminals the most, who can now generate literally infinite content, for infinite amounts of time, because they don't need to verify or prove anything legitimate. And people who are apart of both of these groups are very, very good at sending, to the LLM, prompts that look completely innocent to any kind of guardrail or filter that these companies can devise. The only other use that is probably more profitable is porn. Really.
It’s been said that “Nigerian prince” email scammers intentionally use poor spelling and grammar in order to narrow their funnel at the top by quickly weeding out observant and wary people who are unlikely to fall for it.
By that token, LM-generated content which looks good at a glance but doesn’t hold up to scrutiny seems ideal for scamming. I’d speculate that in the scam content generation workflow, not only is there little penalty to skipping the verification step, but since you intend to push that step onto the target and hope the result is a false positive, subtly wrong hallucination might be not only tolerable but in fact better for its purpose than what a human could produce.
Any business can be run in a shady manner if the human decided. One fully automated business I think could exist and might be useful is apartment/condo rental. I'd pay a business $100 for a proper report on the rentals available in a city that meet a criteria and are amalgamated from all the the various platforms. Doing it yourself (at least in Canada) means creating accounts on a bunch of platforms, and the process is very tedious.
> I'd pay a business $100 for a proper report on the rentals available in a city
I'm curious about things of this nature, where it seems like a case of "this information is important to me and I want accurate results".
But then the talk of automation seems to exclude careful human review of those results, which is needed to stop hallucinations from making their way to customers.
> I'd pay a business $100 for a proper report on the rentals available in a city that meet a criteria and are amalgamated from all the the various platforms.
If this can be fully automated then you can just ask your own agent to do this and wouldn't need a business for it. And agents can already fill out web forms just fine.
Well like most rich guys, I have an assistant, so I don't need or use "agents" - maybe my assistant could learn to use "agents" - but her core competency isn't, nor should it be, learning to use AI agents in any meaningful way. Maybe she could outsource it to someone who got their agents to do it for her for $100.... Same with my little sister who has a 5 year old and a 2 year old and doesn't really know how a computer works never mind what AI agents are.
> Any business can be run in a shady manner if the human decided.
No kidding.
> One fully automated business I think could exist and might be useful is apartment/condo rental.
We're starting strong on the category of businesses that generate no actual value and just scrape an amount of value out of existing transactions that would've happened anyway, i.e., rent-seeking. But good for you, you can now artificially shrink the supply of limited-availability goods in the market, then gate access to them behind a paywall, and you don't even have to do the minimal amount of actual work required to fleece strangers for part of their paycheck while creating no value.
Despite paying rent for an apartment, it’s not rent-seeking. You get a place to live out of it that wouldn’t exist without the owner renting it to you.
Rent-seeking is a very specific economic term where a party inserts themselves into a transaction and takes a cut without providing anything: https://en.wikipedia.org/wiki/Rent-seeking
Being a landlord comes with significant responsibilities and even principal investment risk.
> Despite paying rent for an apartment, it’s not rent-seeking. You get a place to live out of it that wouldn’t exist without the owner renting it to you.
> Rent-seeking is a very specific economic term where a party inserts themselves into a transaction and takes a cut without providing anything: https://en.wikipedia.org/wiki/Rent-seeking
> Being a landlord comes with significant responsibilities and even principal investment risk.
Economist here. Yes, this was a correct use of the term "rent-seeking behavior". It's actually quite funny to see someone try to argue otherwise, when the name was chosen because this is, literally, the textbook example.
It is not the textbook example. The textbook example is regulatory capture where people put in an artificial gate and charge people to cross it.
Everything functioned fine without the gate and nothing was improved by the gate.
An apartment LEASE is literally nothing like that. You’re borrowing something you don’t have and it’s a rivalrous good so other people can’t use it while you are.
Renting (leasing) a car, an apartment, or any other good like that is not rent seeking behavior. No actual economist would argue that because it dilutes the term to something completely meaningless.
A landlord is partially rent-seeking. Yes they provide the service of making sure the apartment is habitable (cough) and so on, but they charge above market price for that. How do I know? I know because I'd do it myself for cheaper if that was an option, but it's not an option because landlords own all the spare apartments. (Why don't I buy one then? They're very expensive because I have to price-match the landlords, who are paying very high prices for the right to rent-seek!)
The market for real estate is basically the market for taxi medallions. It costs something to run a taxi, but there are a limited number of medallions and you can charge well over that cost because you have a medallion, which also makes the medallions very expensive. Until Uber comes along. But you can't just make an illegal apartment without land the same way you can make an illegal taxi without a medallion.
So if anyone accepts your challenge will you move the goalposts and tell them their business isn’t good enough in your point of view? It doesn’t seem like you’re actually interested in dialogue. You also don’t seem to be aware of the definition of rent seeking but that’s an entirely different topic.
I’ll sit out your little experiment because I’m not in the mood for this kind of response. But you may discover that if you turn down the venom a little, qualified people could teach you things like automated business models that are quite ethical and even the definition of rent seeking.
> So if anyone accepts your challenge will you move the goalposts and tell them their business isn’t good enough in your point of view?
It's not a value judgement, it's literally rent-seeking behavior. You're seeking, to rent, property that you own, presumably for a profit. Like come on, it's what the word means.
> You also don’t seem to be aware of the definition of rent seeking but that’s an entirely different topic.
Both my command of the English language and the economist elsewhere in this thread disagree with you, but go off I guess.
> qualified people could teach you things like automated business models that are quite ethical and even the definition of rent seeking.
And yet instead of citing one you went off a tone-policing rant.
My question was quite open-ended. I genuinely didn't expect someone to come in and list the textbook example that an actual economist went on to point out was crap for the exact reason I said, truly. But that's the kind of poetic unawareness that one really can't plan for.
We'll see how this shakes out - to me it is more reminiscent of the boom in bitcoin and associated shitcoins a few years ago, where boosters thought they were going to take over finance and replace our financial institutions in a tech revolution involving NFTs, apps built on blockchain and all money moving to blockchains and everyone not accepting that a revolution was taking place was not going to make it.
That didn't play out quite how the cheerleaders expected (though the value of Bitcoin at least is still high, NFTs and all the actual use-cases for Bitcoin fell through).
I suspect we'll see something similar for LLMs, frankly they're nowhere near good enough for unsupervised use, and if you think they are, good luck to you in building a business on them.
I disagree frankly, as the next wave is clearly fully autonomous businesses.
Considering the disaster of that AI-powered store in San Francisco, I'm skeptical that this could happen in the next wave. Or even the next ocean.
(WSJ article from a few weeks ago stated that the "AI" can't stop ordering candles, and manages the staff so poorly that sometimes there are no employees scheduled for some shifts.)
I'm convinced that one of the top use cases for OpenClaw is orchestrating cold outreach email campaigns, as if there's nothing wrong with using AI to spam people to death. Platforms that enable sending cold emails are taking a sizeable risk that the low engagement of such emails stimulates some worsening inbox deliverability for the rest of their traffic (see [1] - you can't hide just by sending through big tenanted platforms like Amazon).
[1] Every message sent from Amazon SES carries a "Feedback-Id" header that allows Google (and anyone else) to track the Amazon account responsible for the message. The fourth field is an opaque but stable identifier associated with your Amazon account; receivers can and do use this for rate limiting:
https://aws.amazon.com/blogs/messaging-and-targeting/underst...
Technically, you could create many top level Amazon accounts, but if you want to send lots of mail, you must warm up your account. So accounts can be created, but it’s useless if you need to send high volumes of messages.
And cloudflare can actually sell them priority access to pass their bot protection or introduce micropaiments for agents access content. I feel cloudflare is getting a bit scary tbh. It is like your friendly bot net.
Depends on how democratic the Society is. The less so, the more it's a powerful minority at the top. How much say did the average US citizen have in starting this war with Iran? It's not something the current administration ran on to get elected. It's not very popular in the polling.
Yea, I appreciate them protecting it from DDOS. I always viewed them as a responsible company.
To me this feels irresponsible and like it's main goal is to forward autonomous cyber attacks. Which is antithetical to what they do? Maybe I am missing the legitimate use case here, but I can only see this being used for removing responsibility from crime or espionage?
Does anyone know offhand if cloudflare is a department of war contractor? I never looked into it. But this smells funny to me
Somehow the Internet needs biometrics and age verification everywhere but also chat bots can buy property there without too much thought.
Most people don't get DDOSed, and for many of the ones that do, they can just wait it out until the attacker gets tired of burning money. It's very costly and risky for the attacker. Obviously they do occur sometimes - so does murder - but Cloudflare is massively exaggerating the risk of drive-by shootings to make you buy their bulletproof vests.
I was about to say, this type of automated domain purchasing and deployment is a godsend to scammers.
They buy up a bunch of .top, .shop and .xyz domains for $1 each. They spam them out in in all those "USPS tracking" spam text and Facebook fake store ads. The shelf life on these domains is a few days at best before they get shut down, now you can automate domain rotation and not have to pause your spam campaigns.
Are you perchance talking about deSEC? I've also switched to them, and thought that it was too much work to send an email and wait for replies, so I ended up using dummy inboxes for my other, lesser important domains.
Though I guess it's still a good thing they do this? At the time I remember being mildly inconvenienced, but not enough to actually care. I just remember thinking, "How is this nonprofit going to handle all that support volume?".
> Perfect for spammers, scammers and domain squatters, who can now automate their activities even more.
Buy why do this, unless you're in the business of arming both sides of a conflict?
(With a side bonus of designing a defense product in such a way that you reverse much of the ground that was gained with "HTTPS everywhere", to give you centralized cleartext access to much of the Internet traffic.)
Generate hyper personalized ads in any format! Embed them stealthily into virtually any context!
I guarantee thats all big tech is thinking about as the future of LLMs. All this coding stuff is merely a good will investment to buy them time until they can implement this
I have a hobby or we could even say an addiction of having lota of ideas for various kinds of things. Part of the fun is inventing a name, visual identity, logo etc. and donain nane is part of it. Maybe they never reach production but many have. I have tools to automate boring parts like any developer would — and I could essily use something like this.
I kind of agree with you, I would hazard that this is perhaps targeted towards folks (maybe TPMs, small business owners) who are using ai to start a software side business or make a website and have never bought a domain before or configured DNS.
I guess also; something that saves me 20 minutes a few times a year is still nice.
The only use case that pops into my mind is to build a product like Shopify that sets up a store, email, landing page, etc all from one chat-bot interface.
I think this post needs to be put in context, for months now Cloudflare has been releasing products that allow their whole platform to be usable by agents with the main objective of enabling their customers to dynamically write code using Cloudflare, this is just another step.
For example, you can now with Artifacts and Dynamic Workers make a lovable-style SaaS where your customers ask the AI agent to write software for them, the agent can run it in sandboxes with no build step, it can version it with a git-compatible API, and now you can even have it buy a domain for the end customer or set up their own cloudflare account when they want to move to production.
I personally have no use case for creating domains via agents, but some of the other features they're releasing around this area are extremely useful and I've started to ship internal tools for my clients where they are used, like giving them their own mini claude code that only does one thing – one I shipped last week was an agentic interface for Salesforce reports that understands their domain better (and all the undocumented tech debt) than the built-in Salesforce AI does and therefore manages the context better
On the tooling I've made? Feel free to ask here or via email, the Salesforce one was for a fairly large (50k+ employees) company that uses it very extensively but as it often happens they're stuck with a ton of legacy crap. They have some of the AI tooling from SFDC but barely use it (lack of training, or interest) so this solves their immediate problem.
I'm very keen to use their new dynamic workflows (cf's durable execution engine) which would let agents write workflow steps, that way my users can ask an agent to do stuff like "run this report daily and email it to me" and it can work with minimal setup (very basic example, but you get the idea)
It's for founders who don't have lawyers. My co-founder and I are both developers, we used Stripe Atlas to incorporate a C-Corp due to expecting to fundraise <1 year after incorporation. Stripe Atlas generates about 200 pages of legal boilerplate documents with very sane defaults so that your corporate structure, bylaws, IP protections, director indemnity, etc. align well with investor expectations. It helps investors not have to "rules-lawyer" all your corporate records during due-diligence, because their content exactly matches YC's expectations.
-------
I said we made a C-Corp but other founders should default to LLC, which Stripe Atlas can also streamline. An LLC is superior to C-Corp in pretty much every way for any pre-raise founders who don't have an extra $2,000 to >$10,000/year they're willing to part with for higher franchise taxes, "foreign" (different state) corporation registration, CPA's, and additionally lawyers if any investments aren't YC SAFE's (e.g. not YC, Neo, or A16Z SpeedRun).
Also note that for pre-revenue C-Corps, Delaware franchise taxes are scaled against number of shares, not company revenue or # of employees, so you can save some money by forming your company with 1,000,000 shares and then file a "Unanimous action of the board of directors" to increase it to 10,000,000 just before angel/pre-seed/seed round, and potentially save a few hundred dollars on your first year franchise taxes, depending on when you incorporate and raise. But if a few hundred dollars makes a difference to you, incorporating as an LLC instead of a C-Corp is the only defensible decision.
And as always, start your taxes 3-4 months before they're due. If you want a CPA to do them (which you should if you have any revenue), you'll need to retain them way ahead of time for C-Corps. If you're filling tax forms out yourself, you'll want to start at least a month before they're due.
Yes unless you have a very very very good reason it's always best to just file a basic LLC in the state you are a resident of. Only costs a few hundred dollars at most and doesn't really complicate taxes
This is totally false, sorry. Delaware entities are the standard. Delaware corporate law is better understood than any other by a long shot. Dealing with a random non-Delaware LLC is usually a hint that your counterparty is a rube.
Huh? You got any sources on that? I’ve never heard anyone say a non-Delaware LLC is a bad idea. And most sources seem to say incorporating in your home state is usually the right call.
I guess you didn't read my comment. I said unless you have a very very good reason one of which could very well be if you plan on immediately raising VC money. Most companies don't do that and will just end up wasting a lot of time by having a foreign LLC and tons of additional tax issues they have to deal with by having a corporation.
"The minimum tax is $175.00 for corporations using the Authorized Shares method and a minimum tax of $400.00 for corporations using the Assumed Par Value Capital Method." [0]
That is the "save a few hundred dollars" I was talking about. I did get the # of shares threshold wrong, it needs to be <= 5,000.
> They're wrong about the foreign registration -- in California (and I believe most other states), you also need to register foreign LLCs.
Yes, but I was referencing that it often costs more to register a C-Corp than an LLC (depending on the state).
> They're wrong about investments -- SAFEs are very easy for corporations (no lawyers required), but they can't even be used by LLCs. You'll need to convert to a C-Corp.
Yes. Totally agree on all points. This conversion will cost roughly the same as first year taxes, but leaves the option of not doing it if you never get enough revenue to hire employees and don't get funded. And if you do get enough revenue for that, or you get funded by SAFE's, you'll have no issue affording the lawyer+CPA who can do it for you.
As for it being a bad time to deal with that headache, I generally agree. You'd probably want to do that when you reach the point that you feel ready to start fundraising.
> They're wrong about investments -- SAFEs are very easy for corporations (no lawyers required), but they can't even be used by LLCs. LLCs don't have stock, and most boilerplate documents will not work for LLCs.
I miscommunicated on this point: I meant to say if you're not getting funded by SAFE's, you'll need a lawyer, and therefore the "saving money" thing probably isn't particularly relevant and there's no issue filing as a C-Corp.
Boilerplate documents work fine for LLC, and Stripe Atlas helps with this.
> Something about passing losses from an LLC to your personal taxes being a good way to get you audited.
I'm not sure you can do that? Haven't had to deal with it personally (my LLC's were profitable in their first year) but AFAIK capitalization is usually done with post-tax money so I don't see how first-year LLC losses can reduce your personal AGI.
> Something about tax paperwork burden being roughly equal for LLC vs. C-Corp.
I was mainly trying to say that CPA's charge more for C-Corps than for LLCs.
People use agents to deploy sites all the time. Buying a domain is part of that if you want to build a site that's beyond a toy. Allowing agents to do a task isn't just for things you do every day – it's also for things you do rarely and need agents' help. It's not just devs using agents to perform these sort of tasks anymore.
Stripe Atlas makes it massively easier for startups to incorporate in Delaware. This is particularly hard for non-US founders. It solves a real problem. I don't think this part will be done by agents though!
I am watching people who can't code build and deploy dashboards and sites with Claude Code (desktop app - they don't use the CLI), then go cap in hand to developer friends to get it hosted on a domain (rather than some Vercel or whatever URL).
Those people absolutely want to risk letting an agent buy and set up the domain.
This is not necessarily as blindly stupid as you might think. Many of these people know that this workflow is no good for writing code that does anything serious (i.e. storing data for people, taking payments, etc.) but there are a huge number of projects that are just websites, dashboard, data visualisations, etc. with static content and public APIs (Twitter is awash with them) and domains are cheap.
A decent minority of these are even quite cool or interesting.
So a lot of people want to put their vibe-coded weekend project behind a nice domain. Why not?
Let's say they buy a first time discounted $5 domain with a $9000 renewal (could the first renewals be made contractually mandatory?), potentially some other weird terms that the agent agreed to for them.
If I was ill spirited I'd go look at how the agents try to buy and setup juicy traps to milk it as much as I can for the first wave.
I would expect the value of a domain purchase + setup handled by an agent is the highest for people that are not very technical. I'd say that a well-engineered agent will do a better job avoiding botching it than your average non-dev.
If the rest of your deployment flow is via the agent, needing to switch over to a different context and open up a browser and login (or create an account) and buy the domain absolutely is a bump in the road.
Lets remind the purpose of incorporating in Delaware is legal tax evasion, so that we don't not have pensions, health insurance or anything nice, really.
Are you sure you know what you're talking about here?
In the US, regulations on pensions, health insurance etc. are governed by the state that employees physically work in, not by the laws of the state of incorporation.
Please explain. Your comment reveals your lack of understanding of corporate law and the benefits of one state versus the other. And smart companies are going to incorporate in Texas anyway and it has nothing to do with taxes. More to do with corporate governance.
Investors usually expect that non-US founders incorporate in the US, and usually expect Delaware. There are other states that are more friendly to tax avoidance. Delaware is mostly preferred because it's a known quantity with mature regulation. Investors don't want to deal with dozens of different legal regimes, they want the one that they know about.
The primary purpose of incorporating in Delaware is less about taxes and more that Delaware is the "Silicon Valley" of corporate law - incredible concentration of professionals, infrastructure, and intangibles. Any dispute you have will generally be handled better, faster, and cheaper by Delaware courts than they would be anywhere else. I'll quote my good friend who is a startup M&A lawyer: "I'd go so far as to say that it would be managerial malpractice to incorporate anywhere other than Delaware."
Nevada makes it much harder to sue corporate officers when they do malfeasance. Wyoming has tons of privacy perks for the officers (similar to cayman island accounts). “Perks” though also convert into signaling for the intent of the founders.
Uhuh. And in other places, companies are incorporating in Ireland or Luxembourg or other similar tax evasion heavens because of the "predictable legal framework" too. Lol.
Last time (after years of doing it manually every once in a while) I just gave codex an ephemeral restricted Cloudflare API Token / key / whatever, the screenshot, and it set up all the records on its own.
My biggest hesitation with these things is that there is no limit to the possible bill I may receive when the agent goes haywire. Cloudflare doesn’t see this as a problem of course.
> The reason this blog post does not come with any concrete examples how to use this enablement for useful and constructive things tells you something very important - it is a toy and they do not know who and how they will use it.
Every time I come across AI projects and AI integrations (including my previous job where I full-time worked on one), no one was able to show me concrete examples how can I use it for constructive things.
> Buying a domain is not something you have to do daily to require any kind of automation.
Which is arguably unfortunate, as it nudges people towards using centralized services because they simply don't know that they have the option to register one.
For example, why not self-host a single-page party invitation site designed by an agent rather than using Facebook or Instagram?
A lot of what enabled Web 1.0 was how easy it was for an average web user to create his own website.
An average web user got far less technical since, and making a website got harder instead.
Now, if anyone could just ask an AI agent to set up a website, and get a personal page with an e-mail inbox and a domain - all reasonably secure, TLS set up, billing added as +$5 per year to the AI subscription bundle? Maybe that would help some.
Yes, this is exactly my hope too. Many hacker/cypherpunk ideas failed or never reached wide adoption because they were just too complicated for regular people: GPG/web of trust, self-hosting websites and email, having your own custom software for personal tasks…
Instead, everybody ended up using Gmail, iMessage/WhatsApp, and Facebook, and things are as centralized as they can be.
Agents could be a force in breaking that trend. Even if inference stays centralized, the artifacts agents create would not be. Basically the difference between everybody renting from one of a handful apartment building mega corps or being able to hire contractors to build your own things according to your ideas.
And just like there, it’ll probably help a lot to know a bit about how the sausage is made to not be taken advantage of. Also, many people will probably always continue to rent, which is fine. But the possibility of agent competition alone will hopefully keep centralized platforms and SaaS offerings on their toes, which is good for their users.
The problem is not website, the problem is discovery and discovery is on Instagram, TikTok, and social networks. You don't have any incentive to build a website for a regular audience. What you might do is build an audience on a social network and then try to move them to a website.
But at that point you're big enough to build it properly.
You can always follow the POSSE pattern [1] (except for platforms that actively punish links to your own site of course). That way you get both the reach and remain independent in terms of content moderation.
1980s and before - Centralized computers, thin clients & terminals
1990s - Decentralized computing - rich clients running native code locally, data lives locally, making use of the network for necessary communication (e.g. email, IM)
Present day - Entirely centralized data, compute outsourced from companies to 'Public Clouds', bloated clients running JavaScript locally, but mostly without local storage besides cached copies of data.
I suppose it might be one of those things that just oscillate in society, since neither extreme is without frustrations. And as soon as the generation most familiar with the specific downsides of either extreme steps aside, the cycle of history repeats.
This was such a weird mention to see in the article. Stripe Atlas is a service that helps new businesses incorporate and onboard onto Stripe/partner services with some startup credits. It's been around forever, has nothing to do with AI, and is generally a very well-respected service.
So, I have been doing a lot of vibe coding and making projects and got a confident grasp of the AI agent capabilities (primarily a Opus user, Claude code, and have ~10 years of full stack experience).
When I recently started a consulting agency, I found myself reaching into Cloudflare and experienced the same bottlenecks they describe.
Here's where I can see this helping again, in the future, when I know what I want and I want to quickly provision a quick digital presence:
- buy the domain
- provision and deploy to the domain, use Cloudflare as my hoster to serve my bundles (already created the site locally and have MCPs configured to handle this)
- Give it a budget, my "risk" tolerance amount, of say $50
- if something goes wrong, the ~30mins saved is worth that $50 risk (low chance of problem here)
- Yes I'm going to play with this and see if it works out, and if it does, now I can spin up startup websites for small microapps and POCs a bit faster.
People making cooking websites, websites for their garden, etc usually have nowhere to go. A web app who is an agent for a customer will then deploy agents in the backend to deploy the website too.
Basically what one would do manually, you tell one agent to make another agent do it.
It's certainly a great and useful tool. But it's a website maker somewhat in the same way that a Facebook page or Instagram account is a website maker.
AFAIK you can't make a website on SquareSpace and download it to your computer, edit it locally and move it to a different host, etc.
In the past there were actual WYSIWYG editors which let you design your website or CMS theme and then do whatever you wished with it. Artisteer was the pinnacle of this. Then nerds took over with their command lines and Kubernetes.
Imagine if one day people decided that making and editing documents in Word was no longer possible, that they had to be hand coded and command line compiled and linted, and not mix tabs and spaces. That's what happened to website publishing. For no reason at all.
I think that's the real gap. Non-technical people don't want to learn DreamWeaver or SquareSpace's backend. They want to describe what they need and have it just work.
While large social media sites have captured lots of traffic, etc. I've had small websites for a local wargaming club, a very modest blog, etc. for decades requiring little or no technical expertise.
The idea that people who want modest websites need active agentic systems to do that is a really odd take.
Sadly they will be publishing on a web which has no human readers anymore because it’s been crowded out by 5 trillion AI slop gardening websites. And the only visitors will be other AI scraper bots.
Any actual readers will be on platforms which combat the bot spam.
I am curious. Say, cloudflare let's Mr doofus run 4 agents from AI provider X. Those agents go on to create a pyramid scheme, prompted or not(both cases are interesting to me). The agents are running code in infrastructure owned by a third party cloud provider. The law catches up with the bots many years later after people lose a couple million. Who is at fault?
I think cloudflare is in the clear. Mr doofus could argue that the AI company allowed or enabled the crime which they otherwise wouldn't have done. Or Mr doofus could claim his prompts shouldn't have lead to that outcome and that wasn't his intent at all. Making the bots at fault, but not the AI company I guess?
An entire generation of non developers are being onboarded to products built for developers because of AI. Exampl: a very non technical friend has built a bunch of business automations with Claude and a few days ago he just dropped in the conversation that he is using Cloudflare. “For what?” I asked because usually Claude will recommend Cercel before Cloudflare. But in this case, it recommended Cloudflare because of the Tunnels feature. Now he is using it for a lot more than Tunnels. Mostly just lets Claude use it whenever it makes sense.
“To what end?” was the default reaction when Amazon launched S3 in mid 2000s :)
> I've have personally never seen a good example where a cross vendor account provisioning actually working.
That's not what this is though, is it? In other words, isn't the (anti-)pattern you describe an argument in favor of agents setting up your accounts instead?
You can tell your agent to buy the domain at registrar x, manage DNS at y (and maybe configure DDoS protection and CDN), and host your content at z, and if the agent is good enough, you don't even need to understand the details.
You end up with individual credentials for each service, rather than a web of account relationships managed by a single "portal" SaaS.
> I am also not sure who Stripe Atlas for. I am genuinely confused. It is definitely not something a developer will use.
Stripe Atlas has actually been around for quite a while, from like 2016. It's a quick start business platform for startups. HN discussion here: https://news.ycombinator.com/item?id=11166417
So correct in that it's not something intended for developers. It's intended for entrepreneurs (whether or not they are developers)
> The reason this blog post does not come with any concrete examples how to use this enablement for useful and constructive things tells you something very important - it is a toy and they do not know who and how they will use it.
I've been wondering the same thing. What would have stopped me from using a model to talk to their APIs in an autonomous way anyway? Are you going to captcha my bot from running scripts to hit your API end-points? How can you tell its not my automated scripts vs an agent vs me testing?
I remember back in the day Heroku had a huge store of integrations that you could just turn on with a click and they worked like that. You'd get a New Relic account that was tailored to dyno performance and you accessed it via your Heroku dashboard.
It became "the way" a lot of these PaaS systems operated and I'm sure the goal was to get some percentage once you increased your usage from the free tier, which makes sense for the PaaS partner.
For sure, revshare is standard on those partnerships.
Fun(ny) fact: all the companies that started out on Heroku back then are still locked into those Heroku-captive tenant accounts on those partners, because contractually, the partner is not allowed to transition such an account to direct billing. One company I've worked with has had all their infra moved off Heroku for almost a decade, but their Sendgrid account, which has hundreds of subtenants that each have custom domains configured, still can only be logged into via Heroku. They'd have to rebuild that whole thing from scratch (including make all their customers redo DNS validation) to move to a real sendgrid account.
I'm sure Heroku earns Salesforce a really healthy revenue stream based on this.
I don't think there are a lot in the SAAS world. Usually when something quirky and new launches, readers on this website can discern something about useful intent.
Arguably Github, Slack, Twitch, TikTok were basically toys at launch with a lot of people questioning possible market fit.
But there is a difference between those products - and for example - everything that came out of the crypto blockchain scene. This new product by Cloudflare feels more in the latter camp than the former.
Slack, of course, came out of a video game company pivot. Dropbox is famous here because of that one take. The need to allow very intelligent people to be whimsical is maybe best captured by Richard Feynman spinning dinner plates at the cafeteria, which would later lead him to a Nobel Prize for it.
Seasoned, well trained engineers have created hundreds of thousands in billing on AWS through a simple mistakes overnight.
It's immediately reminds me of putting microtransactions into children games on mobile devices - a venue that has been thoroughly explored some 10-15 years ago.
I can't see payment and provisioning as a blocker in any scenario.
This has a potential to create a massive yet very dubious income stream for the company.
Yea its just you and did you generate that with AI ?
Plenty use Cloudflare to ship and this basically negates the need to automate all the logistics. Not really sure what you are rambling on about, you are just mixing totally unrelated group of tools and services to support your "yeah agentic use of cloudflare is bad: which makes no sense.
> Buying a domain is not something you have to do daily to require any kind of automation
I wrote a python client for dnsimple nearly 16 years ago to exactly that. If you can’t think of a reason it’s useful, you may wish to get your agent to buy a domain for some project you have asked to create.
Doesn’t this sum up most of the AI “innovations” we’ve seen shoveled in this bubble?
We constantly see AI thought leaders backpeddling on promises and just spouting general nonsense. Altman originally talked effusively about an era of “abundance”. An abundance of what? It’s a word salad of feel good vibes without any substance.
Sam Altman has gone from claiming AI might cure cancer to shoveling ads and the scope of AI seems to be reduced to mostly be suitable as flawed, imperfect, but mildly useful coding/automation agents that are likely subsidized beyond economic viability, but you can’t point that out because it’s the future!
I'm probably very out of date here, but I thought domains weren't allowed to be purchased programmatically due to misuse, crime, fraud etc. Why is it allowed now just because of agents? This is bonkers to me.
This is simply Cloudflare seeing the possibility of an infinite stream of income... these APIs are cheap to execute for them but are always tied to billing.
Downvoted. Stripe Atlas is a massively successful service that handles 100K+ incorporations annually. It's a key piece of Stripe's ecosystem support. You writing that you're "not sure who it's for" suggests to me that maybe you're out of your depth here as a general matter.
> Buying a domain is not something you have to do daily to require any kind of automation.
Sorry, but no, you totally miss the fact there are domain farms which buy the dropped domains and then offer them up for sale. Bots now use AI to analyze the domain's value and automate the whole process. To be able to let AI buy it as well likely offers a tremendous amount of time saving.
They may be wrong on that particular point, but Cloudflare definitely profits from increased crime as it drives increased sales of Cloudflare's security products. There are rumors they even knowingly help protect DDoS botnets because they benefit from there being more DDoS.
Domain registration is already API driven and has been for decades. The most sophisticated domain name investors (or "domain farms") go as far as to own registrars directly so they have instant access to the registries. Nobody involved in domains would use Cloudflare's product because they already have and have had automations for decades.
For example, DropCatch (NameBright) own over 1,000 different registrars so that they have over 1,000 direct routes to Verisign's .com registry. GName are a new player in the space, approaching 1,000 registrars. The amount these companies spend on their registrar licensing alone is many millions of dollars[1].
Cloudflare's product adds nothing new to the world of domains. Anyone has been able to go to OpenSRS and sign up as a reseller with API access for over 20 years.
If you want an end to end automation solution, you have to automate everything, not just high frequency tasks. It’s not acceptable to just say “oh you can automatically deploy a new site but first you have to register an account and buy a little domain”. The user command is “deploy site, right NOW”.
LLms are an incredibly expensive way to learn that most humans are indeed unoriginal. So armed with an LLM.. yeah great things arent coming out are they.
But at the same time they can't be displaced on a whim either lol.
We just migrated an AI agent from Kimi to GLM and frankly I am surprised by the results. It feels premium.
However, both Kimi and GLM can end up in doom loops so be careful how you use them. Without a proper harness the agent can easily get into some tricky situations with no escape.
We had to develop new heuristics in our cloud harness just because of this but I am really grateful that we did as the platform feels now more robust.
reply