But I shall disagree and take the position that the Keynesian cross should be taught in macro classes.
All of his points are actually well taken, the problem is he makes those points as if it's an either/or choice between AD/AS and the Keynesian cross. Of course it's not. He also makes those points as if you don't have a semester to clean up the Keynesian cross with other intuitions.
I don't have the teaching or the textbook-writing experience that Nick does, I can only say what I did as a TA in a 300-student macro class (which meant weekly review lectures and longer exam review lectures).
Of course the broader set of "I dunno - supply?, demand?" responses came out of instruction with AD/AS but the Keynesian cross was nice to teach too because:
1. They could actually solve it - the math is trivial so you can take non-econ majors that are required to take it and they can work through it.
2. It reinforced the income-equals-expenditure point which is one of the critical differences between macro and micro and for that matter between macro and how people think generally, and
3. It's actually easier to teach AD/AS when you've established my point #2. AD/AS is actually a lot trickier than it first appears. The AD curve does not slope down for the same reasons it slopes down in micro classes. The AS curve does not slope up for the same reasons it slopes up in micro classes. In fact I wish they didn't even call them "supply" and "demand" curves because that makes people get stuck on the micro logic. You need general equilibrium logic to think about the behavior of AD/AS and the Keynesian cross is actually a great tool to teach that.
4. Contrary to what some people have been saying in this discussion, the Keynesian cross is the perfect time to teach about crowding out. Why? Because when you give them plausible numbers to plug in and have them solve you get stupidly high multipliers (and I'm talking about multipliers of four or five, not Landsburg's multipliers). Then you show them a chart that Mark Zandi put together and ask "so why don't we actually get multipliers that are that high? where can we see daylight between what the model says and what happens in the real world?". My (limited) experience is that that tends to be a very fruitful discussion.