Give the prototype a sensible home.
Think about running costs and operations while the system is still small.
The first deployment establishes habits that are easier to shape early: how work is queued, how failures are understood and how costs are tracked.
Separate the request from longer work.
Some tasks fit comfortably inside a web request. Document processing or multi-step analysis may need background execution. Make progress visible and decide what should happen if a job is retried.
Know what failed.
Useful logs connect a request to the services it touched without exposing private inputs unnecessarily. Distinguish a provider outage, an invalid result and a problem in your own application. They need different responses.
Match infrastructure to actual usage.
Choose deployment and storage around the workload and the team's operating capacity. Review idle resources as well as peak demand. Keep a route to change the setup when usage teaches you something new.
A modest, understandable starting point makes it easier to learn what the product really needs.