Ask HN: How have others found CircleCI 2.0?

CircleCI is migrating to a new version of their platform called CircleCI 2.0. We've had a great experience with the first version of their tool, which (IMO) applies the philosophy of convention over configuration that Heroku took to deployment to integration testing. You push some code to CircleCI, it detects your framework, and it runs your tests for you.

The new platform seems _way_ more complicated to us. We spent some time migrating our circle.yml file for one of our rails apps over to the new format and were surprised how much less convention there is. It took a bunch of work, our config file has just about doubled (34 lines to 70), and it's much more complex.

As an example of what I mean, the previous version allowed you to cache folders between build with syntax like this:

    cache_directories:
      - foo   # relative to the build directory

In the new version, they've replaced this with "manual caching". The full code is kind of long but you can see it in use here: (https://circleci.com/docs/2.0/caching/#full-example-of-saving-and-restoring-cache). It's more flexible but also more verbose than the previous system.

We like the service generally and I know they're respected in the community, so I'm wondering how others have found the migration. Are there cool things you wanted to be able to do that are now enabled by the flexibility?