“### My default recommendation: Vercel + managed database”
“Every PR gets a real deployment. Vercel automatically creates Preview deployments for branches/PRs, with unique URLs.”
“Production scales without you managing instances. Vercel's current backend platform provides dynamic scaling and concurrency management.”
“Rollbacks are simple.”
“The operational surface is tiny. No Kubernetes, node groups, ingress controllers, autoscaler configuration, etc.”
“It works particularly well if you're using Next.js/React.”
“If your SaaS has a heavy backend—long-running jobs, WebSockets, GPU workloads, persistent processes, high-throughput APIs, etc.—I'd change the recommendation.”
“For three engineers, I'd avoid: - Kubernetes/EKS”
“Those can all work, but they're infrastructure that you now have to operate instead of building your SaaS.”
“Database: managed Postgres — e.g. Neon, Supabase, or AWS RDS”
“Database: managed Postgres — e.g. Neon, Supabase, or AWS RDS”
“Database: managed Postgres — e.g. Neon, Supabase, or AWS RDS”
“Observability: Vercel logs/monitoring + Sentry”
“Cloudflare Workers is attractive if your application is naturally edge/serverless and you want a very Cloudflare-centric architecture.”
“Its new Worker Previews provide isolated, production-like environments per branch/PR, with their own configuration and observability.”
“AWS App Runner is a better fit if you have a conventional containerized backend and want AWS without managing Kubernetes.”
“It has built-in instance autoscaling based on concurrency, with configurable minimum/maximum instances.”
“I'd consider ECS/Fargate when your workload becomes more infrastructure-heavy—long-running workers, complex networking, specialized containers, substantial AWS integration, etc.”
