Model-Driven Engineering has been promoted for some time as the solution for the main problem software industry is facing, i.e. complexity of software development, by raising the abstraction level and introducing more automation in the process. The promises are many; among them improved software quality by increased traceability between artifacts, early defect detection, reducing manual and error-prone work and including knowledge in generators. However, in our opinion MDE is still in the early adoption phase and to be successfully adopted by industry, it must prove its superiority over other development paradigms and be supported by a rich ecosystem of stable, compatible and standardized tools. It should also not introduce more complexity than it removes. The subject of this paper is the challenges in MDE adoption from our experience of using MDE in real and research projects, where MDE has potential for success and what the key success criteria are.