Then I ordered the book, and during the month it took to arrive (back in the 80s), worked out the themes from the article in programs of my own. When the book came in, it was extremely instructive to compare with how I'd tried to flesh out their ideas -- for example, I'd come up with a scheme to express dataflow pipelines by hooking up coroutines in continuation-passing style, inspired by their CPS-coded pattern-matcher; then the lazy streams in the book were a revelation -- a beautiful solution to the same problem, as a data abstraction instead of a pattern of control. (And then it was interesting to see how you could refactor the different solutions into each other.)
I'm not going to tell you to go about it the same way, but there is a principle there: struggling with a problem yourself before seeing how the masters do it, you're much better prepped to learn from them (and eventually you'll find yourself getting somewhere they didn't). Feynman emphasized this point.
Anyway, book in hand, I worked straight through doing most of the exercises and also related problems that'd occur to me along the way -- anything that piqued my interest. Again, going off on your own tangents is not to be scorned. The concepts or sections to pay particular attention to are the ones that interest you, yourself.
I spent 2-3 months, I guess, taking up all my hacking time. It was totally worth it, though nowadays there's far more competition in easily-available worthwhile programming stuff to study. I've heard of some online study groups for SICP -- don't know if they're active.