Except that the entire article is written around the concept of a one-dimensional pendulum that simply counts movement starts & stops. That is not how activity trackers with 6 motion dimensions sensors work. His one paragraph that obviates the rest of the essay only even considers "two or three accelerometers." It is as if the essay is 10 years old, and it got a minor update a couple years ago. The point is that modern devices to not use a simple method like the author describes-- they are able to trace entire movements of an arm in all dimensions, and they map those movements to activity patterns that align with steps or whatever else you are doing.
Here is a super cool example... my Garmin 910XT counts swim strokes. If it just had a simple two or three accelerometers, it could know that I completed a stroke, but that would be about it. Instead, the watch not only knows when I complete a stroke, but it also knows what kind of stroke it is (it differentiates between back, breast, fly, and free). That is the 6-dimension thing going on with CPU interpretation. And because of this, what it can count is far more accurate and useful than a simple pendulum approach. The same applies to how the AW counts a step.