cto.new is democratizing AI coding the way Gmail democratized email.

They offer frontier models completely free to anyone.

Their bet is simple: AI won't replace developers, it'll make them superhuman. Instead of writing repetitive syntax, developers orchestrate background agents that handle the grunt work while they focus on the big picture.

"We wanted to offer a best in class AI code agent that anyone can get access to completely for free. We think this is the way the industry is heading," says the team.

The response has been overwhelming, over 30,000 developers signed up in the first week, forcing the team to implement an invite system.

Problem

cto.new's platform runs untrusted AI-generated code for thousands of developers.

Every coding task requires spinning up secure, isolated environments to analyze repositories and execute potentially dangerous code safely.

When you're giving away this compute-intensive service for free, you need infrastructure that can rapidly provision secure sandboxes at scale without breaking your economics.

They needed a solution that could handle unpredictable traffic spikes while maintaining the isolation and startup speeds that make their platform viable.

cto.new's platform works by spinning up isolated environments for each coding task. When you connect a repo or ask for code generation, they create secure sandboxes to analyze your codebase and run AI-generated code.

Their initial setup: Firecracker microVMs running on big EC2 metal instances. For early testing with limited users, this worked fine. Firecracker gave them fast startup times and good isolation.

But load testing before launch revealed the problems:

• Scaling meant huge jumps. Adding capacity required full metal instances costing thousands per month, used or not
• Slow, unreliable provisioning. Metal nodes took too long to spin up and were sometimes out of capacity
• No granular scaling: You couldn't scale smoothly with demand
• Operational overhead: Managing metal instances, networking, security, all stuff they didn't want to spend time on

"We had Firecracker VMs running on a really big EC2 instance," says Sudhir Balaji, their technical lead. "That worked well for quite a long time. Then prior to launch, we did load testing, which revealed it might not work so well with a huge spike in traffic."

(Great hindsight given their successful launch!)

The team had tried other platforms before but found them unreliable. One previous solution they'd used simply wasn't consistently online when needed. They needed something that could handle their specific requirements:

• Docker-in-Docker support
• API-driven provisioning
• Pay-per-use pricing instead of big capacity jumps
• Minimal operational overhead

"We're a small team," explains cto Paul Groves. "What we really want to do is focus on shipping features to our users. There's no dedicated DevOps or sysadmin on our team. We are polyglot engineers who want to build product."

Solution

The migration happened faster than expected. Sudhir thought he'd need extensive support calls and custom work. Instead, he got their system working in a couple of days just using our docs.

"Northflank’s docs were very straightforward and basically entirely sufficient for me to replicate what we had done on Firecracker over months in like a couple of days."

Their current setup: For every repo connection, they spin up two Northflank VMs, one for environment setup and configuration, another for code analysis and memory. They're now running thousands of projects and services per day.

Results

cto.new's launch exceeded all expectations. They had to implement an invite system within 24 hours to control demand. Throughout the chaos of handling 30,000+ signups, user feedback, and various product issues, their infrastructure performed without problems.

"Northflank has offered us essentially an out-of-the-box solution that has been basically flawless," says Groves.

"We've been launching for a week. We've been doing a lot of firefighting, chasing down errors and user behaviors on our platform. And Northflank hasn't skipped a beat for us."

The scaling economics made sense too. Instead of paying for entire metal instances regardless of usage, they now pay for resource consumption with per-second billing.

With infrastructure handled, the team can focus on their product differentiation.

cto.new is building orchestration systems that understand project context and suggest what to do next, moving beyond simple prompt-to-code conversion.

The platform flexibility matters for their roadmap too. As they add more integrations and expand capabilities, they need infrastructure that can adapt without forcing architectural changes.

"The fact that the future might be that we want some dedicated hardware, we want to bring our own cloud for this specific thing, but we're essentially integrating with the same company, the same APIs, is incredibly good in terms of technical vendor partnering and future proofing," notes Groves.>

For cto.new, the switch eliminated infrastructure as a concern. They replaced metal instance provisioning, boot times, and operational overhead with a solution that scales automatically.

Now: thousands of daily container deployments, smooth performance during traffic spikes, and an engineering team focused on building AI tools instead of managing infrastructure.

Try cto.new if you want a best-in-class AI coding agent that runs the latest Anthropic and OpenAI models, completely free.