We spent a year building our own edge compute setup. Deployed Node services to AWS CloudFront origins, tried to handle request routing and auth there. Seemed smart at the time.

The reality: we were managing 6+ deployments per region, debugging network issues at 2am, paying way too much for compute we barely used. Auth token validation happened at origin half the time anyway because of cache misses.

Switched to Cloudflare Workers last spring. Dropped to a single codebase, bindings to KV for distributed state, and their cache semantics just work. No more "why is this request going to Oregon" debugging sessions. The pricing is weird (per cpu milliseconds) but we're actually spending less overall.

Only real downside: less control over exactly how requests route. But that's table stakes for not running infrastructure. Their error pages are generic too but whatever.

We're still using CloudFront for static assets because the origin pull is useful, but all the logic that used to live in Lambda@Edge lives in Workers now. Faster deploys, clearer code, fewer pages in my incident runbook.

If you're thinking about edge stuff, just use Workers or Vercel Edge Functions depending on your stack. Don't build this yourself unless you're Netflix.