The key is that the contracts need to be minimal and analyzable, so its ~100 lines of code you can manually analyze. Documented separately all the edge cases, or ideally removed through good design so they simply don't exist. A bunch of problems I've seen in the real world of crypto are annoying edge cases that you can fix with an if statement, or ideally change the design to remove them completely. It's actually a surprisingly hard space because you're trying to make terse, complete and secure code.
In normal web/systems programming you have layers of security; internal services, external services, firewalls, access controls, vpns, security by obscurity, etc etc etc. In crypto you have none of that. Everything is 100% public, the code probably should be published, or else it is trivially decompiled. If there is a problem someone will find it if there is value in finding it.