Hi Meinte and thanks for sharing your views on a topic I’ve been following too — albeit the technology started originally with different name, then renamed a couple of times and today called as LW. For example, you might share some observations from 1996 article “is the game worth the candle”.
I was interested to see what you identified as the main challenges and agree fully with your point #1 that pay-off should be as immediate as possible. Based on customer experiences (collected from peer-reviewed articles) that usually comes quickly after 1–3 weeks. This is considered acceptable time. Naturally if building DSL takes months the immediate pay-off is harder to reach. Interestingly, there is empirical indication that among LWs there is a big difference to the effort and getting pay-off. For example, with Eclipse and its modeling framework, it takes 50 times longer to get DSL tooling than with the fastest. So, instead of 2 weeks it takes 2 years. Could that explain the resistance you mentioned?
Your topic #2 is easy to solve: Sell DSLs only to companies and teams who are keen to deliver results not working hours.
Finally, I agree that integration is important, but having ready integration for just developer infrastructure (Git, SVN, Eclipse, Visual Studio etc.) is not enough as by far majority of modeling language users are non-programmers. So integration must start with the excel sheets etc. and LWs must provide out-of-the box many ways to integrate to existing tools, data and formats.