Dedalus Gives Every AI Agent Its Own Computer, in Under 250ms, on AWS
Dedalus Labs gives every AI agent its own persistent Linux machine, built entirely on AWS, so developers go from idea to production in minutes instead of weeks. | 13 November 2026
AWS Founders Stories: Dedalus | Amazon Web Services
Overview
When Cathy Di, co-founder and CEO of Dedalus Labs, spoke to AI researchers, they all said the same thing: the future is set, and there’s nothing you can do about it. She thought otherwise. Di dropped out of Princeton's computer science program to join Y Combinator's Summer 2025 batch with the idea that every AI agent deserves a real computer, not a fragile sandbox. A persistent, fast environment that behaves like the machine a developer would run themselves. Dedalus Labs builds exactly that, the Dedalus Machine, a cloud-hosted micro virtual machine (VM) that spins up a full Linux environment in under 250 milliseconds and runs entirely on AWS. What used to take an agent developer two weeks of Docker files and infrastructure wrangling now takes five minutes.
Dedalus Labs
Challenge | Building agents on the wrong infrastructure
Production AI agents need somewhere to run. For most teams, that meant stitching together sandbox providers and container configs that were never designed for how agents actually behave. The setup was slow, the environments were brittle, and the developers using them were not happy about it.
Dedalus heard this directly from its own early users. “Most of our users told us they hated their sandbox provider and asked us if we could build a better one,” says Di. The company had started out building agents themselves and ran straight into the same wall. “It's impossible to build agents when the primitive that agents are running on wasn't good enough to begin with.”
For a developer with an idea, that gap between concept and a running agent is where momentum dies. Two weeks of infrastructure work is two weeks not spent building the product that matters.
Opportunity | What if Every Agent Had Its Own Computer?
The plan was simple to explain but hard to build: give every agent a machine of its own, and make the infrastructure disappear. “We think every agent deserves a computer that is persistent and fast,” Di says. Dedalus set out to be as much an infrastructure company as a research lab, with a mission to build the compute substrate for an agent-native world.
These environments have to start almost instantly, hold their state across sleep and wake, and cost little enough to run at the scale agents demand. Meeting all three at once meant pushing cloud primitives harder than most workloads ever would.
Solution | Full Linux Machines, in Under 250 Milliseconds, on AWS
Dedalus built its first VM in two weeks. The cold start then was two to three seconds, which, as Di puts it, “in agent world, that's pretty much eternity.” The team kept optimizing. They got it under 250 milliseconds, fast enough that the infrastructure stops being something anyone thinks about.
The entire platform runs on AWS, and every Dedalus Machine is powered by Amazon EC2. The file system runs on Amazon S3 with Amazon EBS caches, and the control plane is orchestrated on Amazon EKS with Karpenter. A full Linux machine with persistent state, sleep and wake, and per-second billing, delivered in under 250 milliseconds, is a combination the team found no other cloud could match. That’s the whole product in one line, and it only works because the primitives underneath it do.
What took two weeks of Docker files, YAML configurations, and infrastructure wrangling now takes five minutes. Dedalus makes AWS primitives accessible to an entirely new class of builders and agent developers who could not provision this kind of infrastructure on their own.
Building on AWS also means a developer never hits a ceiling. Every Dedalus Machine runs on the same environment from the first agent to the millionth, so the same VM that spins up in five minutes scales with no re-architecting as demand grows.
Outcome | Infrastructure You Never Have to Think About
The measure of Dedalus is what the developer stops noticing. “In this future, infrastructure shouldn't be felt,” Di says. “Most people would never know that their agent is running on a Dedalus machine. All they know is that their agent is fast, it's on call whenever they want it to be, and that it's really cheap.”
The market is responding. Dedalus closed an $11M seed round in October 2025, co-led by Kindred Ventures and Saga Ventures, with 20 investors including E14 Fund and Emergence Equity Management. The company launched out of Y Combinator's S25 batch, and says demand for Dedalus Machines is ramping fast.
The team is still pushing the number down. “We're still trying to get it down faster,” Di says. For the developers building on Dedalus, the payoff is already here: the barrier between an idea and a production agent is measured in minutes, not weeks.
Get Started | Build Your Next Agent on AWS
Cathy Di went from a napkin sketch to a platform that gives every AI agent its own computer, built on AWS because agent builders need infrastructure that is fast, persistent, and invisible.
If you're a founder building the next generation of AI agents, you can go from idea to production on the same infrastructure Cathy built on. Explore the AI for Startups hub to start building today.
Did you find what you were looking for today?
Let us know so we can improve the quality of the content on our pages