Modern software infrastructure has accumulated a lot of operational complexity.
It's common to see relatively small applications deployed behind Kubernetes clusters, ingress controllers, distributed configuration systems, and dozens of supporting services before a single request reaches the actual business logic.
Those technologies solve real problems at scale, but for many startups, indie builders, and small engineering teams, they often introduce more operational overhead than product value. Every additional component increases the failure surface, adds cognitive load, and slows down iteration.
Lately, I've been moving back toward simpler architectures with single-binary services written in Go or Rust, paired with embedded databases like SQLite and LSMDB, a lightweight embedded LSM-tree database I've been building.
The deployment model is refreshingly simple. A single executable with minimal runtime dependencies, straightforward backups through filesystem snapshots, and observability without an entire ecosystem of supporting infrastructure.
Simplicity isn't a compromise. It's an architectural choice. Complexity should be introduced only when the product genuinely demands it, not because modern infrastructure makes it easy to add.