Steps: 1. Create a new language in which the problem is trivially expressed. 2. Solve the problem in the new language.
Lisp / Scheme are ideal for this.
Example: solve some sort of puzzle game. Create a data structure that represents the board, a game piece and a move. Operators upon these. Given a board and a legal move, return a new board with the move applied. Given a board, give all of the possible legal moves (for a particular player if this a multi-player game instead of a puzzle).
Once you have that language it becomes easy to use your favorite off the shelf search algorithms. Depth first. Breadth first. A*. Etc.
But, this is probably not the same reason to create a "language" as the article. And the notion of "language" is quite different.
Doesn't a language have to be turing complete to be a language?
It could well be a cultural thing: for some people, "creating a new language" is a serious undertaking, but for others (especially those exposed to Lisp and Scheme), creating a new language is something you might do if you're curious about something. At the end of the day, creating a language just isn't all that hard to give it so much thought.