Jake Cooper, founder of Railway, discusses the company's journey from a slow-growth PaaS to an agent-native cloud platform. He shares insights on building bare-metal data centers, the critical role of primitives for AI agents (versioning, forking, observability), and why Railway is uniquely positioned to serve the next generation of software deployment, where agents increasingly generate and manage infrastructure.
Summarized by Podsumo
Railway's growth exploded after a long slow start; they now add 100,000 users a week with a lean team of 35 people and over 3 million users.
Jake emphasizes 'going deep' into the stack, including kernel patches for performance, and building bare-metal data centers with a 3-month payback period vs. cloud.
The core thesis for the agent-native cloud: agents need the same primitives as humans (version control, forking, observability) but at 1,000x scale and speed.
Railway's CLI is becoming more important than the canvas GUI; agents can now provision and modify their own infrastructure via the CLI.
Jake critiques temporal for its complexity in modeling workflows, suggesting most teams should build simpler workflow engines rather than adopting a 'jet engine' they can't maintain.
"If we copy somebody else's homework, we lose. You have to have a core thesis about why does this business need to go and exist at this point in time."
"The way people will build and deploy software is moving from assembly to C to C++ to JavaScript to now words. Agent-native is the dominant species."
"We always want to be waiting on intelligence, not compute. If you're waiting on compute, there's a bottleneck that needs to be destroyed."