Sprites are a beautiful place to make an abstraction.
But I have never used a buggier infrastructure product in 30 years of development than when I tried sprites. Data loss left and right, sprites going into unconnectable zombie mode. Couldn't use even load snapshots because half the system seemed to think the sprite was fine, and half seemed to think it was dead. I gave up after two weeks of losing stuff over lunch or overnight or in the middle of working. I vividly remember scrolling through my terminal history, copy/pasting out work I could find to save from a dead sprite. I would guess that half or more of the sprites I launched had some kind of problem.
Hope they get things stable. The concept is great.
I operate an app that orchestrates code running on a bunch of Sprites. A few months ago it was an exceptionally buggy service, frequently just fully broken or losing data (had to get support involved to recover one that got into a fully broken state), but from what I've seen at least they've become much more reliable in the past two months or so.
I evaluated Fly.io itself several years ago and found it unacceptably buggy as well for any real use case at the time, but I kept an eye on it and today I operate a few production workloads on it that are very stable and reliable—I figured the same would be true for Sprites and it seems to be taking shape that way.
Can second this. I had an enterprise client that was interested in using these for fast development etc. Wasn’t my project or the reason I was there but I was curious so I kept up with it.
Eventually they abandoned the company but stuck with the concept by going with another company. Complaints were the same, extremely buggy interface, dataloss, but also terrible support. That may have been there fault though for not using their actual domain for the test period (and not mentioning that they were an F200 or something).
I won't claim strong enthusiasm for Sprites, but in my experience (albeit way back in 2021) Fly's support was absolutely top-notch, up to and including Kurt personally working with me to help debug our deployments. This was as a small business only billing a few hundred a month.
> (and not mentioning that they were an F200 or something).
I am not a F200 company or anything, just a normal guy but I have the opinion that I personally like the companies who treat normal accounts and the general experience/support. I have just felt like as if its a good litmus test depending on things. What are your opinion on things and was the enterprise client opening up a normal account intentionally similar to line of thinking or for other reasons?
I had some conversations with the lead on the project and his reasoning was something along the lines of “we’ve been suckered into using something with sweetheart attention before only to have the rug pulled when we’re ‘reliant’”.
I also think he might have not wanted the attention that putting the company name on the account might get.
Were those good reasons? Honestly I don’t know, but apparently the experience of using the product spoke for itself.
> “we’ve been suckered into using something with sweetheart attention before only to have the rug pulled when we’re ‘reliant’”.
It makes me feel like this is like the corporate form of love-bombing perhaps, oof, I think that two reasons for this could be that most companies would want to do anything for contracts with large companies, (especially within B2B?), of course I feel as if some additional attention is a good point but excess of anything is bad as you have said that they learnt from. Its good to know that they learnt from it though :-D and perhaps other companies could also learn from it as I believe that wisdom comes from learning from other people's mistakes as well.
Another reason is that companies like to really show about other large companies using it. I have heard when a company literally said (users of this product also use)... and then proceeds to show the names of recognizable companies like Lovable/OpenAI/Anthropic/Google. (that too very sneakily)
There is also the part of used by engineers at (Big tech) etc. but I think I remember cases where the engineer doesn't use it as part of their day job at Big tech but rather on the side and there are so many engineers and the amount of use could vary just as well (because otherwise the company could just say "used at Google/Microsoft etc." itself.)
I found that part silly but perhaps my heart still tries to understand other people and my explaination of it is perhaps that with all the new companies, one of the filters that people select for trust is seeing what other large companies are. This makes us, say, transfer the trust we have from the brand of these large companies (Google/Microsoft/Lovable etc.) and into smaller brands.
Most people are going under identity crisis right now because of recent LLM advancements. This post is a good example that shows that it's not only happening at the individual level, but also on the company/organization level.
Is it still worth building products or companies that can be one-shotted by AI? Probably not.
One interesting consequence is that this force everyone to be more ambitious and do something bigger that's impossible before.
I hope more people are working on something that can always bring net positive to humanity even if there are hundreds of people working on the same thing, like clean energy.
As an elixir dev fly was a company I very much wanted to succeed but I twice had to leave them because they just never found the right balance between cool engineering and ops.
For a long time they would have global outages where the status page would list everything as green and the only reason you'd know something was wrong was the forum threads. Their response was that they were too busy fixing the issues to update the status board. Right.
Then once they finally started updating the board they would go several hours between updates with no indication of what was happing "Region x is down" then nothing, again "too busy fixing the issue".
When they came out with paid support I signed up immediately only to find that it gave me an email address to contact support with fast responses promised but most of the time no one was monitoring it. I'd report large outages and not get back responses until the next day and on a few occasions I'd get back a response several days later being like "Can you explain the problem you're seeing?" No mate it was an outage several days ago.
This would all just be annoying customer service if it was a rare occurrence but the reality of using them was that I was dealing with major outages almost monthly at points. If you scroll back through my post history here there were brief periods where stability would improve and I'd get optimistic about them only for it to crash back down to regular outages again days after I posted.
I eventually had to move everything off them to self hosted again which is a pain but far less of a pain since my uptime is dramatically better(!) and when something breaks I know what it is.
I still want fly to succeed but if you're going to host peoples stuff you need to accept responsibility and move budget to ops. Otherwise just kill the hosting entirely and try to become another HashiCorp.
> We’re launching a new iteration of Sprites, and focusing the company on them and the problem they solve. And I’m tagging in Scott Johnston as CEO.
Hard for me to see this as anything other than fly.io committing suicide. AI sandboxes are a crowded product space, basically a complete commodity at this point, and the new CEO will likely focus on profit to the detriment of creative vision.
The AI sandbox play also confuses me because (aside from the fractional pay-as-you-use billing) it's trivially implemented on top of any VM/container as a service provider. Given the reduced importance of data retention (the containers are disposable) and ability to redo work trivially you can even go a step further and have your agent spin it up on bare metal. Turns out agents spend most of their time waiting on GPUs so with memory dedup on your own similar workloads you can run "hundreds" of them on pretty minimal hardware.
I also don't see how the cloud-for-agents will break the AWS monopoly: AWS already is the cloud for agents! As the article notes all the pain of using AWS is trivialized by agents (and IaC).
To me the value now is in building the right tools for agents to make better decisions, not fighting for basis points on hardware margin in the orchestration or model serving layer.
Sprites are clearly the future. Are there plans to give it a web UI? I haven't minded ssh'ing in, but I've wanted to share this with my wife who would use a sprite on her iPad. Basically, ChatGPT but with a full sandboxed Linux environment that can host stuff is a pretty compelling offering.
Check out exe.dev for basically what you are describing. Also remember these aren't really sandboxes per se, they're just computers/microvms. Exe also has a neat auth/https proxy that also does TLS/certs they have lots of cool tools and integrations.
I don't think these sprites are worth betting the company on, but I guess that will be for the new CEO to decide.
> A Sprite is a hardware-isolated execution environment for arbitrary code: a persistent Linux computer. Whether it's an AI agent like Claude Code or a binary your user just uploaded, Sprites are the simplest answer for "where should I run a blob of code".
I think eventually this isolation may be built in to Claude code / codex, or provided by AI companies themselves.
Sprites are about more than just providing an environment for Claude Code et al to run code in. They're also useful for anyone who's building their own product that needs an inexpensive way to execute unsafe code.
Why are you so sure "DX" is the right way to compete with commodity full-stack hosting? Have you had Claude deploy on AWS recently? It one-shots deployments just fine.
I have a question regarding sprites but when executing unsafe code, because by definition its unsafe, what happens if the code violates terms of service of fly/sprites. Would only that particular sprite be removed or would it escalate to whole account level ban and the risks attached to it or would they take a look at repeated offenders.
from fly's terms of service Section 1.4(d) and section 9.2 respectively
> Customer will not... "sublicense any of Customer's rights under this Agreement, or otherwise use the Fly.io Services for the benefit of a third party."
> "Either party may terminate this Agreement for no reason or any reason upon written notice to the other party, effective immediately at the end of the then current Subscription Term."
I know git worktrees solve similar problem, but you can use sprites for quickly spinning up your instance and run your service and hand it over to coding agent to improve a feature.
you can run multiple parallel agents to do same and choose one of them as a final result, and you are not constrained by number of ports or your CPU (because some agents are CPU hungry at the moment)
Think about even more use cases, add to that RAM shortage is not going away yet, you can run 100s of agents in those sprites
Yuk I don't want to rent my appartment from my drug dealer. I love I can say thanks for the code Claude now fuck you I am deploying it (waggles finger in westerly direction...) there.
Sprites seems like a solid lifestyle business for a few friends, not a startup that has taken funding.
As a developer, I use docker/podman. Works great.
The new massive market that will open up, where grandma and grandpa are making apps, will be taken up by Lovable (and the like). Non-devs will not be using sprites.
You’re capturing some small percentage of existing developers.
Saw this on my RSS feed and was surprised to see it hasn't been discussed yet on HN - fly.io used to be on the front page pretty regularly with some pretty big comment threads, but I had to go back over a year to find anything from them with more than a handful of comments. I guess they've become less relevant on here since they stopped allowing new users to host apps for free?
What do you mean? We ran a big post about Litestream Writeable VFS just a couple months ago, and we've had several big posts over the last 12 months, including the Sprite launch. We got pretty busy, and there's a lot going on, and I am a bottleneck on big posts going up on the blog. We have a couple months of technical catching up to do. But it's not like we've vanished off HN.
Ahh ok I just did a search on hn.algolia.com for fly.io and nothing came back with 10+ comments in the past year (until I posted this). But now I'm realizing they just don't include the url in the search, even though it says they do
Right, just go to the front page and click on the linked domain on this story, it'll take you to all the previous submissions from our blog. It's been a light summer on blog posts, to be sure! But that's nothing new.
Oh ok i thought they charged a flat fee to use the platform now but maybe that was a temporary thing. My account is a few years old so i wouldn't know, i still have a few apps running on the smallest vm size and it's always under $5
Good discussion on the HN thread about that article. To be honest, based on the latest models and agentic engineering, that blog post is definitely right. It's not 2023, 2024, or 2025 anymore, agents have gotten a lot better.
I think the effect was less about it being correct or not, but instead more about if fly.io would treat humans as first class users compared to agents or not in the future.
The message is fine, but it was a weird choice to have it on fly.io instead of a personal blog
It's just marketing, going with the times of lots of vibe coders now suddenly wanting to put something on the Internet. I wouldn't read too much into it as a fundamental philosophy.
It's right in the point it brings up that modern LLMs are serious engineering and quite a bit capable at many tasks. It's not correct on the rest where Ptacek appears dismissive because 'he doesn't wanna hear' about copyright and other things, maybe we chalk it up to an upset guy at people he believes and maybe knows are wrong and ignore the rest of the spiel, but I won't.
Yeah, I think that post from June 2025 has aged pretty well. The Coding Agent revolution hadn't really happened at that point (Claude Code was only ~4 months old) so the paraphrased message "this stuff is actually useful for programming, saying it's all just stochastic parrots misses the mark" was still non-obvious to a whole lot of people.
Honestly, the biggest problem with that article was the title. People read as far as "are nuts", got offended and refused to engage with the content of the piece.
This opening paragraph does a much better job of capturing the core argument:
> Some of the smartest people I know share a bone-deep belief that AI is a fad — the next iteration of NFT mania. I’ve been reluctant to push back on them, because, well, they’re smarter than me. But their arguments are unserious, and worth confronting. Extraordinarily talented people are doing work that LLMs already do better, out of spite.
The title was the point of the post. I have acquaintances who are still sore at me about it. But the problem wasn't that people disagreed with me about LLMs; the problem was that we were locked in an unproductive and static equilibrium of people talking past each other and hedging their way out of any direct, empirical confrontation over what was happening.
It's easy to forget how different early 2025 was to today. You would generally feel like you were in an inverse Emperor's New Clothes situation if you believed what your own eyes told you about how LLMs wrote code. That's not the case anymore; you hardly ever see people talking about (or, really, extrapolating wildly about) hallucination today.
(For a lot of reasons I think a lot about how that post was received in 2025 and what the world is like today. "Vindicated" doesn't capture the feeling.)
I still find it incredibly troubling. Not because the substance was wrong, but because a person of your caliber has stooped to the level of writing provocative headlines and language to attract attention. I strongly believe world-class technologists are supposed to be above that sort of thing. Leadership matters: not just thought leadership, but attitudinal leadership.
Thomas--as insufferable a windbag and terminally-online personality as he is--was not wrong in writing that post.
It's also hard to find a sufficiently anodyne way of writing that particular post, because quite frankly such a large a vocal contingent of the tech sphere is just so goddamned wrong about what was obviously happening in front of their lying eyes.
To use a coarse analogy, consider how difficult it would be to find a nice way of writing an article that says "folks, gravity is real and everybody saying that dropping an apple isn't going to have predictable results is wrong".
"You are all fucking idiots" is sometimes a reasonable starting point.
> consider how difficult it would be to find a nice way of writing an article
Writing is often difficult. That's what makes good writing so distinctive and valuable. Just because something is hard doesn't mean it's not worth putting in the effort.
> "You are all fucking idiots" is sometimes a reasonable starting point.
No, it isn't. I think we'll just have to agree to disagree on that.
> Honestly, the biggest problem with that article was the title.
Pretty sure the content of a company's blog post being overtly arrogant, abrasive, dismissive and generally dickish was also a problem.
> Meanwhile, software developers spot code fragments seemingly lifted from public repositories on Github and lose their shit. What about the licensing? If you’re a lawyer, I defer. But if you’re a software developer playing this card? Cut me a little slack as I ask you to shove this concern up your ass.
> Yeah, we get it. You don’t believe in IPR. Then shut the fuck up about IPR. Reap the whirlwind.
I would like to take this moment to apologize to you, personally, for not giving a shit about "AGI". I wish I could tell you I've come to see the error of my ways, but I continue not to care about it.
this was amazing post. i enjoyed reading it, it was relevant to some work stuff i was battling with. the linked articles at the end were amazing too. i had tonnes of respect for fly.io and yourself because of these posts. technically informative, clear, engaging. great stuff.
the llm post? complete polar opposite.
> I would like to take this moment to apologize to you, personally, for not giving a shit about "AGI". I wish I could tell you I've come to see the error of my ways, but I continue not to care about it.
thanks for the snark, and kinda proving my previous point about the tone of the article that was supposed to be a personal blog post and then ended up on a company bloghttps://news.ycombinator.com/item?id=49052217
I'm not being snarky. I genuinely do not care about AGI.
It's weird to watch people write as if I'm trying to back away from that article or hedge or whatever. I expect to wrap my career up at some point and look at that post as one of the better and more valuable things I've ever written.
Indeed, it's not worth engaging with AI skeptics anymore today, it's just a waste of time. If they don't see the value then they probably never will, and that's not my problem.
This is what I mean about the static equilibrium in 2025. That's what everyone was doing: just writing the other side off. That post of mine did a lot of things, but it did not do that.
It's not 2023, 2024, or 2025, but LLMs still suck at programming. Just like every other time AI enthusiasts have claimed the latest and greatest is so much better than what came before.
the way certain people at that company speak to others (e.g. that post) is why they will never count me as a customer. although im sure they dont care.
The CEO didn't write that article. I did. I was going to run it on my personal blog, and he asked me to run it there. Be mad at him for the right reason.
It's probably getting less discussion because there's less user energy, the existing VM / database product has clearly not been a top focus for a while and in any case that business is becoming even more of a tough commodity for the reasons the post points out.
> fly.io used to be on the front page pretty regularly with some pretty big comment threads
But that doesn't mean they are not doing stuff. It's just HN doing HN things. At the moment we're into "AI is eating the world alive" and other topics are in the background.
I think they became less relevant, at least amongst human users, when they hit a particularly rough patch with awful reliability issues a couple of years back. At least that's when my own social circle and loose constellation of developer acquaintances started saying "we should probably avoid Fly.io".
No one talks about new software anymore. All the conversation is meta about LLMs that write software instead. This is a dead career, I’m just waiting for the LLMs to start writing code in binary.
I've played around with a few sprites and it's a very interesting concept. You ssh in and then it feels very much like a local development environment, except the dev server you start up and run locally is actually out there and accessible. It's a great way to vibe code a low traffic, personal site. I wasn't super clear on just how much scale it actually can reach, though. Maybe I missed it but I didn't really know how beefy of a server it was on or if I could scale up. But maybe a big scale is not really the intention of this product?
Yeah that seems like a weird way to end an article that includes this paragraph:
> Fly Machines and our Platform As A Service features aren’t going anywhere. But Sprites was the product of a tiny skeleton crew inside of Fly.io[†], and now it isn’t.
Well, there's doing both in terms of focus and effort, basically treating them as peer products. And then there is doing both in terms of "keeping the lights on for one, and focusing on the other".
Without any special insight into fly.io, I'd bet the situation is the latter. You keep the lights and fix security issues for the existing infra (which incidentally gives you money out of cashflow to invest in new efforts).
This really sucks. I not only like fly a lot, I have two production apps; thousands in annual creds up for renewal soon; and some mildly embarrassing discussions about migrations to have.
Whenever I start to use fly.io they have outages all the time. And also I think they are a bit pricey too. I stopped using them, they have unreliable infra.
Macs let you disable suspend just like (or possibly better) than any other operating system. The point being made is that it's the wrong tool for the job, precisely because of the user desired compromises made to create a great mobile computer.
Yeah exactly, I don't understand the trend of it to be honest. Buying a VPS (even a low end vps like 500mb ram or 1GB could be good) and you can use maki.sh (written in rust) as compared to opencode/claude-code etc. which I have had OOM errors with at low end vps's.
You can ssh into your vps's from anywhere and they could really work 24/7 without having to worry about battery or the whole opening mac up while moving because an agent is running thing.
tl;dr - I really like your products, but just don’t have the cognitive resources for many learn->try->decide->use loops, and I don’t think I’m alone here. “Claude, use the sprite skill” is the vector of adoption which matches my mean available energy.
—-
Some feedback on the pivot, from a user. Reading through the recent blog entries, there’s some Very Neat Stuff mentioned that integrates Sprites into tools like Hermes. I excitedly dug right in, to find that they aren’t anything usable, and the Hermes PR needs work. More salient though, that months-old Hermes PR *needs attention* if this is now your #1 business priority. Same goes for umpteen other integrations to shove the Sprites model (which is unique) into the popular agent harnesses.
I tried Sprites when sandboxes were peak exciting, and “yes I can ssh into this and run Claude” was useful but not groundbreaking. From the docs I’ve just read, I was holding it wrong, not taking advantage of its differentiating features. Great, ^H ooh look, mythos, shiny … and now it’s almost August.
And so honestly - despite my appreciation of Fly - with the firehose of Cool Shit, I don’t have the time or the inclination to learn how to hold it. Instead I want a skill/plugin/integration/llms.txt/sumerian tablet that I can give to my LLM to do that “multi-user auth and don’t worry about it you’ve got checkpoints” stuff for me. And I need this before I make any choices about investing in this thing over Bubblewrap or a VPS or the 100 other frictionless zero-knowledge sandboxy things that I am using already. Maybe you have this, and this is PEBKAC. Wouldn’t be the first time.
This is probably great news. Using Fly at different points along the way has been one of the worst experiences I’ve had in terms of their persistent reliability issues and how they go dark on addressing those issues at times.
Whatever culture Kurt built for Fly, it was the wrong one.
Well, fuck. ... we raised a fuckload of money and didn’t need more. And if we fail, we fail with our full asses.
I have seen this language from other companies to which I say "no". Being professional means being able to do good work, being polite, without lying, cheating, or stealing.
I do not want them to fail yet I would never use them despite liking their product. A few other companies behave the same way.
But I have never used a buggier infrastructure product in 30 years of development than when I tried sprites. Data loss left and right, sprites going into unconnectable zombie mode. Couldn't use even load snapshots because half the system seemed to think the sprite was fine, and half seemed to think it was dead. I gave up after two weeks of losing stuff over lunch or overnight or in the middle of working. I vividly remember scrolling through my terminal history, copy/pasting out work I could find to save from a dead sprite. I would guess that half or more of the sprites I launched had some kind of problem.
Hope they get things stable. The concept is great.
I evaluated Fly.io itself several years ago and found it unacceptably buggy as well for any real use case at the time, but I kept an eye on it and today I operate a few production workloads on it that are very stable and reliable—I figured the same would be true for Sprites and it seems to be taking shape that way.
Eventually they abandoned the company but stuck with the concept by going with another company. Complaints were the same, extremely buggy interface, dataloss, but also terrible support. That may have been there fault though for not using their actual domain for the test period (and not mentioning that they were an F200 or something).
Either way I wasn’t privy to the actual details, just that it was one of their gripes.
I am not a F200 company or anything, just a normal guy but I have the opinion that I personally like the companies who treat normal accounts and the general experience/support. I have just felt like as if its a good litmus test depending on things. What are your opinion on things and was the enterprise client opening up a normal account intentionally similar to line of thinking or for other reasons?
I also think he might have not wanted the attention that putting the company name on the account might get.
Were those good reasons? Honestly I don’t know, but apparently the experience of using the product spoke for itself.
It makes me feel like this is like the corporate form of love-bombing perhaps, oof, I think that two reasons for this could be that most companies would want to do anything for contracts with large companies, (especially within B2B?), of course I feel as if some additional attention is a good point but excess of anything is bad as you have said that they learnt from. Its good to know that they learnt from it though :-D and perhaps other companies could also learn from it as I believe that wisdom comes from learning from other people's mistakes as well.
Another reason is that companies like to really show about other large companies using it. I have heard when a company literally said (users of this product also use)... and then proceeds to show the names of recognizable companies like Lovable/OpenAI/Anthropic/Google. (that too very sneakily)
There is also the part of used by engineers at (Big tech) etc. but I think I remember cases where the engineer doesn't use it as part of their day job at Big tech but rather on the side and there are so many engineers and the amount of use could vary just as well (because otherwise the company could just say "used at Google/Microsoft etc." itself.)
I found that part silly but perhaps my heart still tries to understand other people and my explaination of it is perhaps that with all the new companies, one of the filters that people select for trust is seeing what other large companies are. This makes us, say, transfer the trust we have from the brand of these large companies (Google/Microsoft/Lovable etc.) and into smaller brands.
Is it still worth building products or companies that can be one-shotted by AI? Probably not.
One interesting consequence is that this force everyone to be more ambitious and do something bigger that's impossible before.
I hope more people are working on something that can always bring net positive to humanity even if there are hundreds of people working on the same thing, like clean energy.
For a long time they would have global outages where the status page would list everything as green and the only reason you'd know something was wrong was the forum threads. Their response was that they were too busy fixing the issues to update the status board. Right.
Then once they finally started updating the board they would go several hours between updates with no indication of what was happing "Region x is down" then nothing, again "too busy fixing the issue".
When they came out with paid support I signed up immediately only to find that it gave me an email address to contact support with fast responses promised but most of the time no one was monitoring it. I'd report large outages and not get back responses until the next day and on a few occasions I'd get back a response several days later being like "Can you explain the problem you're seeing?" No mate it was an outage several days ago.
This would all just be annoying customer service if it was a rare occurrence but the reality of using them was that I was dealing with major outages almost monthly at points. If you scroll back through my post history here there were brief periods where stability would improve and I'd get optimistic about them only for it to crash back down to regular outages again days after I posted.
I eventually had to move everything off them to self hosted again which is a pain but far less of a pain since my uptime is dramatically better(!) and when something breaks I know what it is.
I still want fly to succeed but if you're going to host peoples stuff you need to accept responsibility and move budget to ops. Otherwise just kill the hosting entirely and try to become another HashiCorp.
Hard for me to see this as anything other than fly.io committing suicide. AI sandboxes are a crowded product space, basically a complete commodity at this point, and the new CEO will likely focus on profit to the detriment of creative vision.
Happy to be proven wrong though.
I also don't see how the cloud-for-agents will break the AWS monopoly: AWS already is the cloud for agents! As the article notes all the pain of using AWS is trivialized by agents (and IaC).
To me the value now is in building the right tools for agents to make better decisions, not fighting for basis points on hardware margin in the orchestration or model serving layer.
> A Sprite is a hardware-isolated execution environment for arbitrary code: a persistent Linux computer. Whether it's an AI agent like Claude Code or a binary your user just uploaded, Sprites are the simplest answer for "where should I run a blob of code".
I think eventually this isolation may be built in to Claude code / codex, or provided by AI companies themselves.
You got claude to deploy. Cool.
I clicked a button in my PaaS and I have an oncall responder and incident manager ready to go.
I have a question regarding sprites but when executing unsafe code, because by definition its unsafe, what happens if the code violates terms of service of fly/sprites. Would only that particular sprite be removed or would it escalate to whole account level ban and the risks attached to it or would they take a look at repeated offenders.
from fly's terms of service Section 1.4(d) and section 9.2 respectively
> Customer will not... "sublicense any of Customer's rights under this Agreement, or otherwise use the Fly.io Services for the benefit of a third party."
> "Either party may terminate this Agreement for no reason or any reason upon written notice to the other party, effective immediately at the end of the then current Subscription Term."
I know git worktrees solve similar problem, but you can use sprites for quickly spinning up your instance and run your service and hand it over to coding agent to improve a feature.
you can run multiple parallel agents to do same and choose one of them as a final result, and you are not constrained by number of ports or your CPU (because some agents are CPU hungry at the moment)
Think about even more use cases, add to that RAM shortage is not going away yet, you can run 100s of agents in those sprites
As a developer, I use docker/podman. Works great.
The new massive market that will open up, where grandma and grandpa are making apps, will be taken up by Lovable (and the like). Non-devs will not be using sprites.
You’re capturing some small percentage of existing developers.
https://news.ycombinator.com/from?site=fly.io
https://news.ycombinator.com/from?site=fly.io
[1] https://fly.io/blog/youre-all-nuts/
The message is fine, but it was a weird choice to have it on fly.io instead of a personal blog
Honestly, the biggest problem with that article was the title. People read as far as "are nuts", got offended and refused to engage with the content of the piece.
This opening paragraph does a much better job of capturing the core argument:
> Some of the smartest people I know share a bone-deep belief that AI is a fad — the next iteration of NFT mania. I’ve been reluctant to push back on them, because, well, they’re smarter than me. But their arguments are unserious, and worth confronting. Extraordinarily talented people are doing work that LLMs already do better, out of spite.
It's easy to forget how different early 2025 was to today. You would generally feel like you were in an inverse Emperor's New Clothes situation if you believed what your own eyes told you about how LLMs wrote code. That's not the case anymore; you hardly ever see people talking about (or, really, extrapolating wildly about) hallucination today.
(For a lot of reasons I think a lot about how that post was received in 2025 and what the world is like today. "Vindicated" doesn't capture the feeling.)
I still find it incredibly troubling. Not because the substance was wrong, but because a person of your caliber has stooped to the level of writing provocative headlines and language to attract attention. I strongly believe world-class technologists are supposed to be above that sort of thing. Leadership matters: not just thought leadership, but attitudinal leadership.
It's also hard to find a sufficiently anodyne way of writing that particular post, because quite frankly such a large a vocal contingent of the tech sphere is just so goddamned wrong about what was obviously happening in front of their lying eyes.
To use a coarse analogy, consider how difficult it would be to find a nice way of writing an article that says "folks, gravity is real and everybody saying that dropping an apple isn't going to have predictable results is wrong".
"You are all fucking idiots" is sometimes a reasonable starting point.
Writing is often difficult. That's what makes good writing so distinctive and valuable. Just because something is hard doesn't mean it's not worth putting in the effort.
> "You are all fucking idiots" is sometimes a reasonable starting point.
No, it isn't. I think we'll just have to agree to disagree on that.
Pretty sure the content of a company's blog post being overtly arrogant, abrasive, dismissive and generally dickish was also a problem.
> Meanwhile, software developers spot code fragments seemingly lifted from public repositories on Github and lose their shit. What about the licensing? If you’re a lawyer, I defer. But if you’re a software developer playing this card? Cut me a little slack as I ask you to shove this concern up your ass.
> Yeah, we get it. You don’t believe in IPR. Then shut the fuck up about IPR. Reap the whirlwind.
> but it’ll never be AGI -- I don’t give a shit.
this was amazing post. i enjoyed reading it, it was relevant to some work stuff i was battling with. the linked articles at the end were amazing too. i had tonnes of respect for fly.io and yourself because of these posts. technically informative, clear, engaging. great stuff.
the llm post? complete polar opposite.
> I would like to take this moment to apologize to you, personally, for not giving a shit about "AGI". I wish I could tell you I've come to see the error of my ways, but I continue not to care about it.
thanks for the snark, and kinda proving my previous point about the tone of the article that was supposed to be a personal blog post and then ended up on a company blog https://news.ycombinator.com/item?id=49052217
It's weird to watch people write as if I'm trying to back away from that article or hedge or whatever. I expect to wrap my career up at some point and look at that post as one of the better and more valuable things I've ever written.
"I would like to take this moment to apologize to you, personally, for not giving a shit about "AGI"."
appears to be dripping with snark
They really don't. You're clearly the intended audience for https://fly.io/blog/youre-all-nuts/ !
These people will be eating their hat soon if they aren't already.
But that doesn't mean they are not doing stuff. It's just HN doing HN things. At the moment we're into "AI is eating the world alive" and other topics are in the background.
So are the prior products going to be shut down? If not, then when? if not when, then why not?
> Fly Machines and our Platform As A Service features aren’t going anywhere. But Sprites was the product of a tiny skeleton crew inside of Fly.io[†], and now it isn’t.
So it's both indeed.
Without any special insight into fly.io, I'd bet the situation is the latter. You keep the lights and fix security issues for the existing infra (which incidentally gives you money out of cashflow to invest in new efforts).
There's a world outside Apple.
You can ssh into your vps's from anywhere and they could really work 24/7 without having to worry about battery or the whole opening mac up while moving because an agent is running thing.
Some feedback on the pivot, from a user. Reading through the recent blog entries, there’s some Very Neat Stuff mentioned that integrates Sprites into tools like Hermes. I excitedly dug right in, to find that they aren’t anything usable, and the Hermes PR needs work. More salient though, that months-old Hermes PR *needs attention* if this is now your #1 business priority. Same goes for umpteen other integrations to shove the Sprites model (which is unique) into the popular agent harnesses.
I tried Sprites when sandboxes were peak exciting, and “yes I can ssh into this and run Claude” was useful but not groundbreaking. From the docs I’ve just read, I was holding it wrong, not taking advantage of its differentiating features. Great, ^H ooh look, mythos, shiny … and now it’s almost August.
And so honestly - despite my appreciation of Fly - with the firehose of Cool Shit, I don’t have the time or the inclination to learn how to hold it. Instead I want a skill/plugin/integration/llms.txt/sumerian tablet that I can give to my LLM to do that “multi-user auth and don’t worry about it you’ve got checkpoints” stuff for me. And I need this before I make any choices about investing in this thing over Bubblewrap or a VPS or the 100 other frictionless zero-knowledge sandboxy things that I am using already. Maybe you have this, and this is PEBKAC. Wouldn’t be the first time.
Whatever culture Kurt built for Fly, it was the wrong one.
Random 1990s nostalgia unlock.
I have seen this language from other companies to which I say "no". Being professional means being able to do good work, being polite, without lying, cheating, or stealing.
I do not want them to fail yet I would never use them despite liking their product. A few other companies behave the same way.