He is right. At least at some level.

The most productive coders are those that know the body of prior art and are expert a decomposing their current problem into chunks that have been done, and chunks that need some "new art".

And of course the new stuff must be put together in such a way it is usable off-the-shelf the next time a similar situation is encountered. If you find yourself doing something remotely similar to something you've already done, you should be looking for a way to take your prior work, abstract it up a level, and make it into prior art.