Code And Let Live

Source: fly.io Date: 2026-01-15 URL: https://fly.io/blog/code-and-let-live/

Summary

Essay by Kurt Mackey arguing that AI agents fundamentally want persistent, durable computers rather than the stateless sandboxes that current agent execution environments provide. The problem: sandboxes waste resources rebuilding environments and force awkward workarounds (S3, Redis) to maintain state. The solution: Fly’s Sprites — VMs that boot in 1-2 seconds, persist state, support checkpointing, auto-idle for cost efficiency, and scale to hundreds without container overhead. The vision: applications won’t need professional developer gatekeeping once users can iterate directly with AI on persistent machines.

Implications

Machines API as agent-runtime substrate / competition with Vercel/Render. This is the clearest statement of Fly’s agent-runtime thesis: persistent VMs beat ephemeral sandboxes for agent workloads. It directly challenges the serverless-per-invocation model (Vercel Edge Functions, AWS Lambda) for AI agent use cases. The “durable storage + fast boot + auto-idle” combination is the specific affordance Fly is building that competitors lack. If this thesis is correct — and the “Our Best Customers Are Now Robots” data point suggests it is — then the agent hosting market will look very different from the serverless market it’s adjacent to. Sprites are the bet on what that difference is.