How do you eat a whale? One bite at a time

We're doing it in chunks. Routing and models are the weirdest part so those generally have to have dedicated stages.

We just released multi-step zaps which addresses the "complex" part of complex integrations - bi-directional is tricky to do in a generic way, but we'll figure something out I'm sure!

We do poll for changes. Rate limiting is less of an issue than you'd think - we do have some generic retry and error handling that takes care of it.

We have a crazy good logging and monitoring stack as well as tons of good algorithms to see when trouble is brewing. We often know an API is down before the service announces it on Twitter or their status page.