Trial and error (tinkering) has a really positive asymmetry: the upside is potentially unlimited (discovering something genuinely good) while the downside is bounded (you simply don't lock in the gain and move on), making it superior to top-down design.