The paper considers "selling to a no-regret buyer": repeatedly selling to a buyer that learns over time using a no-regret algorithm. It follows up on (Braverman et al, Best Paper at ACM EC 2018), and obtains several interesting improvements. While the reviewers largely agree on the significance of the contributions, some of the reviewers are very concerned with presentation. While they agree all their concerns can be addressed in a revision, they insist these concerns *must* be addressed if the paper is accepted. While there is a large spread in the scores, there was a thorough discussion among the reviewers and the AC. This paper is closely related to Submission #908, with some overlap in the co-authors, but is sufficiently different. This relation should be discussed carefully. As per one of the comments, the claims re extracting full surplus may need to calibrated a little more carefully.