
founder @limrun (YC P26) - live in Replit, Rork, Momentic and more. emeritus maintainer @crossplane_io
San Francisco, CA
Joined July 2009
- Tweets1.3K
- Following631
- Followers951
- Likes5.4K
We couldn't hold it any longer!
If you have a mobile app, we should talk.
If you built software factories, we should ABSOLUTELY talk!!
Limrun (@limrun) provides the infrastructure to leverage your favorite cloud agents for mobile development too. Today, we provide remote XCode, IOS & Android Simulators to Replit, Rork, Momentic AI, and many more.
Congrats on the launch, @muvaff!
ycombinator.com/launches/QSu…
for better or worse, my fable 5.2 and your fable 5.2 aren’t the same
this is very much aligned with the way we are thinking about using AI for all development. guidance is still necessary. architecture needs taste. agents doing work at all costs means that some costs are too high. intent and attention is all you need.
slopmanifesto.com
4
352
yes, the things that actually matter still require human attention so better optimize to increase the throughput of that attention
software factories feel like a temporary problem. the models will learn to build them for you
the more interesting question to me is how to build a workspace where agents can collaborate with you
6
406
You have to get Daybreak, then ARTEX and then real hackers like @calif_io to try to take your system apart. Lots of time and money, yes, but it’s tablestakes now.
Hugops to everyone involved
Every time I see a company launching 10 different things and failing at 12 of them
Software got a lot easier if you know what you wanna build, and a lot harder if you don’t.
With clarity, you can straight-shot to the solution. Without it, you’ll ship a pile of junk and spend a lot of money doing it. Pre-AI era you would ship nothing, which is an improvement.
3
477
lol
NVIDIA H100: $1.50/hr
NVIDIA H200: $2.99/hr
….
….
Single Expo Simulator: $6/hr (limited concurrency)
NVIDIA B300: $7.10/hr
As anyone with serious scale knows, sandboxes are dead. One call may need 64GB ram, the other may need Mac. You can’t confine agents to a machine.
We proved this infra-as-a-tool-call model for mobile development @limrun at consumer scale and we’re just getting started.
I was wrong about agent sandboxes.
In my article I said pattern 1, where the agent runs on your infra and only calls a sandbox for tools, doesn't work at scale. I moved everything to pattern 2 and put the whole agent inside a VM.
At the time that made sense but today I'd pick pattern 1.
For your own personal agent, a coding agent, or a small project, isolating the whole agent in a VM is still a great default. I would keep doing it.
At real multi-tenant scale it falls apart.
There will be more agents than humans. Every human gets one computer. Giving every agent its own VM running 24/7 means billions of CPU-hungry machines, and we're already short on compute.
Most of what an agent does doesn't need a machine at all. It thinks, calls APIs, reads pages, and waits. Nobody needs a 4 CPU VM sitting idle so an agent can run a simple JS script once an hour.
What I'd build now:
> the agent loop runs on shared infra that scales
> it gets a sandbox only when it has to run untrusted code
> sandboxes spin up in seconds and die right after
> zero state on the machine, everything lives outside it
VMs are a tool the agent calls, but they shouldn't be where it lives.
As Google has been saying for a decade, the problem with this approach is that it provides a false sense of security. You just have to assume breach and protect internal services accordingly with zero trust.
if you’re working for a company in cybersecurity, you should be evaluating your options to abstract your companies services from the internet
what i mean is: why allow an unauthenticated anonymous agent of an attacker to interact with your systems?
consider your architectural levers now, you’re going to need them
2
11
802
When I first met him, I said it’s harder to secure macOS and he had one big laugh. “It’s been a shitshow in Linux last couple of months”
You have much less flexibility with Apple but it’s miles ahead of Linux in terms of security.
Calif CEO @XorNinja reveals how his team hacked an iPhone and maintained root access even after rebooting or updating the device, challenging one of the most common pieces of iPhone security advice:
"In software security, the iPhone is like the holy grail of vulnerability exploitation."
"Since Apple introduced MIE, which is a hardware mitigation mechanism, since last year, nobody actually has published any public iPhone exploit. So we really, really wanted to hack the iPhone for that reason."
"We managed to get root access to the iPhone and even persistent access."
"In the past people said one of the best practices for iPhone security is to reboot your phone after a while because all the malware would go away."
"But we show that there's ways to persist our attacks even if you reboot or you upgrade your iPhone."
@calif_io
4
493
They escaped from macOS Virtualization.framework, again. That's why we work with @calif_io
Slides for my OffensiveCON - Japan '26 presentation 🇯🇵
blacktop.github.io/presentat…
1
3
497
Funny to have to say it, but yeah, we read every line of code. SLA is no joke.
I don't want our infra providers vibe coding, generating giant PR's, skipping reviews, the current meta
We pay for uptime, service, responsible adults doing things
We want our customers to expect the same from us, so we move slower to do things well
The goal is earned trust
5
795
Just hold your vendors to a much higher standard for security and you’re good to go. It’s just distraction to try to build everything in-house while competitors are building features.
narrative violation: preferences will swing back from DIY vibe-coded software to super secure industry-leading SaaS solutions.
while it's super easy to build anything, endlessly protecting and securing it is super hard...and teams want to sleep at night.
4
428
Many are saying this
In under a week, @instabox_sh and @limrun moved 100% of my dev and testing from my MacBook to the cloud.
It’s at least doubled my ability to multitask: long-running tasks keep going after I close my laptop, and parallel work no longer turns my fanless MacBook Air into a space heater.
7
464
I did lots of GPU debugging for our Android product in C code without ever learning the language, and that was last year. It doesn’t really matter
I still can't believe people are rewriting software into Rust for performance reasons without learning any Rust.
I'm optimizing a Go program right now and using Agents to do the work, but I have to steer them a lot or they get stuck in local minima, but I can only steer them because I have 10 years of doing performance work in Go.
If I was optimizing a Rust program I'd have to just keep saying: "make it faster" over and over again until it got stuck somewhere dumb.
7
538
brain surgery by LLMs
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
2
10
754
People are waking up.
Agents don’t have a compute problem, they have an orchestration problem and @limrun has been building on this idea for the longest time for the highest scale cloud agents, aka vibecoding platforms.
Exactly. We see customers saying they built Limrun in-house only to realize they have at least one engineer maintaining it, can’t scale even internally and also pay crazy Mac infra costs.
It’s fun for the engineers with tokens but terrible for the business.
Replying to @ajkemps
To be direct about it: every company that considers themselves a “tech” company is currently building a much worse version of something they could buy off the shelf, and furthermore they often have several teams competing on this front
Then they launch it after 3-6 months of building in a silo and can’t get their colleagues to use it for entirely predictable reasons
Coding agents and their adjacent products are among the most competitive markets of all time. The benefits of a highly competitive market accrue to customers by virtue of this intense competition among suppliers. Econ 101
Is the internal team working on your agent platform really keeping pace with the market? Even if they put in the same hours (they don’t), they don’t have the breadth of customers and hard contact with the market to get the feedback necessary to make the product any good
Big corporates that call their devs “IT” don’t delude themselves in this way. They buy the best, recognize their relative advantage (vendors will build almost anything for you because the competitive pressures demand it), and get back to work
1
10
655




















