[btilly]

TDD works for some things, not so much for others. If you know what you want to build, and how it will work, TDD makes it more likely that you will build it correctly, and that the programmatic interfaces will be pleasant to work with. If you don't know how to make the thing you want to build work, TDD is not going to help you. At all.

As an illustration see http://devgrind.com/2007/04/25/how-to-not-solve-a-sudoku/ for an amusing example of TDD failing horribly for a simple algorithm problem.

[nahname]

TDD helps with design problems, not algorithm problems.

[willvarfar]

Yet the luminaries of TDD actually imagined it might be able to evolve good algorithms...

You might undersell it; they oversold it.