Solo founders: what's the one infrastructure decision you'd make differently?

I'll go first.

I built Sharpread, a forensic SEC filing analysis tool, on a standard in-process rate limiter for managing requests to the SEC EDGAR database. It worked perfectly in development and in early production.

Then I hit a wall. Vercel's serverless architecture spins up multiple Lambda instances concurrently. Each instance had its own rate limiter. They had no idea the others existed. Under any real traffic spike, the instances collectively hammered EDGAR well above the rate limit while each individual instance thought it was behaving perfectly.

The fix was rebuilding the rate limiter on Upstash Redis: a single shared atomic bucket that all instances hit together. One distributed token bucket, enforced at the union rather than per instance.

The lesson: Anything that needs to be true across your entire fleet cannot live in memory on a single instance. I knew this in theory. I learned it in production.

What is yours?