For a solo builder, the traditional server is a tax you pay forever: patching, scaling, monitoring, a monthly bill whether anyone shows up or not. For most of what I ship, I've stopped paying it. I run on Cloudflare Workers — code that executes at the edge, close to the user, with no server to babysit.
Makmur Motor (a Denpasar dealership's showroom + inventory CMS) runs on Workers with D1 and KV. Venmail (a privacy-first temporary inbox) runs on Workers. This portfolio — including its contact form and a Workers-AI chat assistant — is a Worker serving static assets. Three different products, one operational model.
What the edge actually buys you
- No servers to manage. There's no VM to keep alive, no OS to patch, no region to choose. The code runs in hundreds of locations automatically.
- Latency is low by default because the code runs near the visitor, not in one distant data centre.
- The bill rounds to zero at low traffic. A dealership site or a niche tool doesn't need to cost real money to keep online — that changes which projects are even worth shipping.
The storage primitives make it a real platform, not just a CDN: D1 (SQLite at the edge) for relational data, KV for fast key-value reads, R2 for files, and Workers AI for inference right where the request lands.
The trade-offs I plan around
The edge is not a normal server, and pretending otherwise causes pain:
- No long-lived connections or big background jobs. A Worker is short-lived and request-scoped. Anything long-running belongs in a queue or a durable object, not a "just keep it running" process.
- D1 is SQLite, not Postgres. It's excellent for edge-native apps, but when a product needs heavy relational work, row-level security and complex queries, I reach for Postgres/Supabase instead. Right tool, right job.
- Cold, distributed state means you design for eventual consistency where it's fine and strong consistency (a transaction) only where money or correctness demands it.
The honest summary
The edge isn't magic and it isn't right for everything — a data-heavy system of record still wants Postgres. But for the long tail of real products a small team ships, "no servers, low latency, near-zero idle cost" is a genuinely better default. It's why I can run several live products at once without a server bill or a pager.
Top comments (0)