The manual testing really isn't very hard... where it relies on the same logic I use existing logic. Where it's new logic, particularly new UI, testing really can't be automated. Where it's not UI but server logic or DB queries, it's pretty explicit and tested six ways to Sunday before committing. I think the only true unit tests I ever wrote, if you can call them that, were to test pseudorandom behavior across distributions of things like character spawning in games, where player testing wouldn't be sufficient to spot anomalies. For something like that where you're going to have a billion variants, it makes sense to test for oddities on some subset of a million or two. In other words, tests to make sure that output fit a certain curve. Not that logic itself worked or didn't work, because if I can think of a test to write where it might fail, I'd already have fixed it not to fail under that test.