Given this blog post: Turn And Face The Strange · The Fly Blog
You’ve said that you don’t believe in the apps platform anymore, that sprites are the focus of the company now, and that you don’t intend to half ass two things.
So can we get an honest estimate of how long until you start winding it down? Or some kind of public commitment that it’s not going to happen? Because right now it feels like a huge risk to continue with Fly.
He also said “I quit” - so hopefully his successor doesn’t have the same opinion. 
There’s many companies that can do multiple things very well… don’t think it’s too much to ask for Machines, Apps and Sprites to co-exist. 
machines are staying!
Fly Machines and our Platform As A Service features aren’t going anywhere
Let’s hope they stick with it.
Not only aren’t Machines going anywhere, but also we have been hard at work on improving reliability of Machines, or Sprites, and just our platform in general while getting rid of some technical debt we have been accumulating in the past. I’m sure you’ll hear about these either as Fresh Produces, or, on our Infra Log (ideally as Fresh Produces because Infra Logs happen after incidents 
).
“Computers for agents” does not just mean Sprites. Machines are also a huge part of it. They are what got us to here today; and that’s definitely a reason not to wind them down, but instead continue improving them for the future.
The same was written about GPU: We Were Wrong About GPUs · The Fly Blog
In there it was noted that: “GPUs aren’t going anywhere … we’re not getting rid of them.”
Then about a year later: GPU migration - "Fly.io GPUs will be deprecated as of July 31, 2026"
So at least there’s a bit of non-confidence-inspiring history.
that’s a good callout. the difference is that GPUs had very few users, and GPUs cost us a lot of operational effort (and money). fly machines have many, many users, and are growing every day; keeping them running, and improving them, is a better use of our time.
if everyone stops using fly machines one day, I can’t promise we won’t get rid of them then. but as long as the interest to use them exists, and it very much seems to, we still have a good reason to keep them around 
in my experience, startups can do 1 (or often 0) things well. You all have pretty explicitly said that going forward that thing is gonna be sprites. Super bummed, but it is what it is. So I suspect that it’s less that people won’t use them and more you will have choices: spend finite eng time on sprites, or spend on the existing platform.
I hope I’m wrong, but I’m thinking through a migration. If that wasn’t your intention behind that post, maybe an exec should say something to that end. Ideally with a statement about being able to refund credits if the machine platform is EOL’d too…
@earl asked for an exec to say something, so: hi 
(I joined to run engineering).
Machines are not getting wound down. There is no timeline and no internal doc with a sunset date in it.
The blog post said we’re focusing the company on Sprites and the problem they solve. I can see how that could be misinterpreted to mean that we’re ditching Machines. What the blog post didn’t say, and what I’ll be clear about now, is that Machines are part of that solution. “Computers for agents” is not just a synonym for “Sprites.” A very large share of the agent workloads we run today are running on Machines, put there by people who built something that works and have zero interest in rewriting it. Deprecating the platform those customers live on would be an act of self-harm, and we’re not that creative.
Re: the discussion over whether engineering time is finite and Sprites will eat all of it, that’s a reasonable thing to believe about a startup. Here’s what’s different for us right now: I joined as our VP of Engineering, and one of the first things I did was stand up a team whose entire job is platform reliability. We still think “everyone owns reliability,” but now on top of that, we’re dedicating actual people to the work of making the platform you’re already on more boring. @PeterCxy wasn’t being diplomatic upthread when he mentioned some investments we could make (he may have called it technical debt, but that’s a post for another day). That’s the team’s backlog, and you’ll see it as it ships.
On GPUs: fair callout, and @lillian 's answer was the right one. GPUs had a small number of users and an operational cost wildly out of proportion to that number. That was a genuine case of not being able to do two things well, so we chose. Machines are the inverse on every axis we’d use to make that call. They have a lot of users (not bragging, it’s just true), and they’re pointed at the same customers Sprites are. This isn’t two unique things, it’s the same thing in two shapes.
If you want to go click on something rather than take my word for it, @lubien shipped a Fly Machines API connector for Sprites this morning. Point your agent at it and it can drive the Machines API without you handing your Fly tokens to whatever’s in your dependency tree. That is not the kind of thing you build on top of a platform you’re planning to turn off.
On refunding credits if Machines are ever EOL’d: we’re not going to invent a refund policy in a forum reply for an event that isn’t happening. I do understand why you asked, and what I’ll say instead is that when we’ve had to make hard calls before, we’ve looked for the answer that doesn’t wreck people. Accident forgiveness exists because someone ran up a bill they never meant to and we decided that eating that compute cost ourselves beat being “right.” We are not in the business of leaving people high and dry.
Anyway, if you need to migrate, migrate. We’re not going to be weird about it. But you’d be leaving a platform we’re currently spending money and headcount to make better.
Thanks for the replies here, I really appreciate them. I’ve updated the post title to be a little less frustrated and dramatic based on your answers. I really hope that machines stay a core, important product at Fly. There’s not much like it out there and it’s an incredible offering. I’m choosing to be cautiously optimistic.
Also, thank you for dedicating some folks to reliability. That’s the only hesitation I’ve had recommending Fly to others, which has been a shame because I’m otherwise a huge fan of it.
Machines are certainly not going anywhere. 