How are the latest neural-machine-translation advancements changing professional workflows?

Which specific improvements in fluency or terminology consistency have you noticed with the newest MT engines, and how are you adapting your post-edit processes?