Modern database applications often change/evolve their schema, which is a necessary yet often an operation described as "dicey" or "dangerous" due to insufficient support in existing systems. More often than not, application developers find themselves using ad hoc schema evolution features that are provided by DBMS vendors as "patches" to an existing system that is not designed with transactional schema evolution in mind. This leads to various issues such as missing features, poor performance and corner cases that require manual intervention by the application developers. In this talk, we revisit the problem and solve the problem with a new data-definition-as-modification (DDaM) approach that makes online transactional schema evolution a first-class citizen in the database engine level. DDaM fits nicely with MVCC and allows us to seamlessly model schema evolution as table modification operations without the aforementioned drawbacks. We will then describe the Tesseract system, a realization of DDaM with optimizations on a modern multi-versioned main memory OLTP engine, and discuss future directions in this line of work.