Writing a test that fails is a subset of what you can do. Pharo (and other Smalltalks) support writing traditional test cases and running them in a test runner. However you can also:

  - write snippets in a workspace, as vincie mentioned.  
  - write code against an API that doesn't exist, run it, have it fail, and then fill in the API and implementation.  
  - write stubs that fail automatically, and then fill them out (similar to above).

The difference is, when things fail:

  - you get to play with the live objects in a work space.  
  - fill in missing implementation.  
  - fix any errors.

And then much of the time continue execution from the point of failure. No need to restart anything. I simply could not appreciate the variety of options available in a Smalltalk environment until I used it for some fun projects over a few months.

Be warned, you might hate your so called modern environments once you get back to them. For Objective-C coders there is hope: http://injectionforxcode.com

Test your code in a Workspace before editing your classes. Depends on the complexity of the change you are making of course. If it is complex, clone the image and test on the clone.