Case Study
How Asteroid Powers Full-Stack AI Workforces With Daytona
130k+
sandboxes provisioned per month with Daytona
10x
growth in sandbox throughput in only four months
500x
increase in agent execution volume
Martian is an AI lab applying interpretability to build safer and more effective agentic LLM systems. Backed by leading VC firms such as NEA and General Catalyst, its research has advanced the understanding and performance of these systems.
Headquarters
San Francisco, CA
Industry
Software Development
Department
Artificial Intelligence Engineering
Key Features
Sandbox Creation Speed Infrastructure Scale Isolation and Security
Learn how this AI browser workforce platform partnered with Daytona to replace fragile in-house infrastructure with a managed runtime that effortlessly provisions isolated, ephemeral sandboxes for agent-driven workflows.
Daytona transformed what we could do overnight. We went from being capped at a handful of agents to confidently scaling to tens of thousands of executions per month.
Joe Hewett
Co-founder and CTO at Asteroid
01 -- CHALLENGE
Scaling Ephemeral Agent Environments Demanded a Reliable Infrastructure Partner
Asteroid was founded to help today's leading healthcare and insurance organizations automate mission-critical browser workflows. Its AI agents autonomously generate code, navigate complex portals, and execute repetitive data entry workflows that used to require dedicated teams. But following an intensive YC accelerator program and a resulting spike in interest from enterprise clients, Asteroid's existing infrastructure was being pushed to the limit.
Joe Hewett, Co-founder and CTO at Asteroid, witnessed this double-edged growth firsthand. At the time, Asteroid relied on a homegrown agent execution runtime hosted on self-managed infrastructure. While the system handled scheduling and task execution, it couldn’t reliably ship arbitrary code in isolation, especially at the scale Asteroid’s clients demanded.
The challenge extended well beyond reliability. With agents handling sensitive data and interacting with legacy Electronic Health Record (EHR) systems running exclusively on private Windows servers, each execution required its own VM with bulletproof isolation and full compliance with healthcare data-handling standards. Additionally, Asteroid’s enterprise customers often need to process massive backlogs (sometimes spinning up 5,000 executions in a single day), meaning its infrastructure had to scale instantly without falling over.
Joe knew that orchestrating VMs across hundreds of thousands of monthly runs would become a significant operational bottleneck that his seven-person team couldn’t reasonably take on in-house. So, he sought a managed runtime platform that could effortlessly handle secure VM orchestration at scale. That search ultimately led him to Daytona.
People underestimate how hard it is to deploy virtual machines at scale. Daytona does it exceptionally well, especially the Windows snapshotting.
Joe Hewett
Co-founder and CTO at Asteroid
02 -- SOLUTION
A Managed Runtime Platform That Delivers Isolated, Ephemeral Sandboxes at Scale
Joe’s team slotted Daytona in and migrated their entire runtime layer in only a few weeks. The platform solved the code isolation challenge immediately, giving every agent its own full VM with secure, ephemeral networking.
Following integration, Daytona provided the instant elasticity Asteroid required. When a customer triggers a workflow, Asteroid spins up a Daytona machine, the agent completes the task, and the machine is immediately killed. This ephemeral architecture, combined with Daytona’s isolated networking and per-sandbox permissions, keeps sensitive data contained during every run, satisfying Asteroid’s strict compliance requirements.
When enterprise customers need to process thousands of claims or form-fills in a single day, Joe’s team relies on Daytona’s pre-warmed machine cache to absorb the burst. Asteroid provides custom snapshots pre-populated with the exact programs their agents need. Daytona keeps these images warm, enabling Asteroid to provision fresh environments in under 200 milliseconds. This rapid boot time ensures sandbox creation never bottlenecks the application, even when running thousands of parallel workflows.
Daytona also addressed the Windows challenge that Asteroid’s healthcare clients introduced. By providing early access to Windows VM snapshots, Daytona enabled Asteroid’s agents to interact directly with legacy desktop applications. That capability cracked the healthcare use case wide open, helping Asteroid automate slow, manual EHR processes that previously required human intervention on local hospital networks.
Daytona lets us spin up a fresh VM for every workflow, do the work, and tear it down. That level of isolation and scalability is something we could never have built ourselves.
Joe Hewett
Co-founder and CTO at Asteroid
03 -- RESULT
Asteroid Scales Agent Execution Volume by 500x With Daytona
With Daytona, Asteroid replaced its homegrown infrastructure with a managed runtime platform that enabled consistent deployment, scaling, and monitoring of agent workflows. Now, Joe’s team focuses entirely on improving their core agent technology, while Daytona handles the elastic compute layer in the background.
130k+ sandboxes provisioned per month with Daytona
10x growth in sandbox throughput in only four months
500x increase in agent execution volume
Moving forward, Joe and his team plan to scale their parallel execution capacity even further, running thousands of concurrent workflows to serve their growing client base.
Daytona gave us early access, helped us diagnose issues, and treated us like a much bigger customer than we were. That willingness to invest in us early made this growth possible.
Joe Hewett
Co-founder and CTO at Asteroid
