I'd also say start with the fixed. If you don't, you'll likely be providing him with some basic shader(s) that represent the FFP anyways, atleast while he's learning how to send data TO the pipeline. Then once he understands that, you can show him how to customize the pipeline.

I really don't see how there's much need at all to learn the fixed-function pipeline these days. There's nothing you can do with it that you can't do with shaders, and it's not even present in newer API's like D3D10 and XNA. Working with shaders also removes much of the "magic" of the basic lighting and texturing calculations, since its all right there up-front in the shader code.

Bleh! Can we put a stop to this question now?

Learning fixed function graphics can be useful if you are explicitly targeting one of the following platforms in your development:

* Cell Phones/PDAs
* Nintendo DS or PSP
* Last-Gen Game Consoles
* Graphics hardware lower than the GeForce FX series (don't recall the ATI eqiv.)
* Non-shader capable graphics devices

If you plan on programming for NOTHING BUT one of those devices, then I suppose there's no good reason to learn shaders. However, if you have any interest whatsoever in recent graphics methods, newer game consoles, or game development on a screen bigger than 4 inches then let me make this perfectly clear: YOU NEED TO KNOW SHADERS! They drive everything about the modern graphics market today, and starting with DirectX 10 (and OpenGL3, whenever it comes out) fixed function will no longer even be an option.

Basically: Fixed function is going to be a waste of your time unless you KNOW that you will need it because of your target platform. Even then, though, learning shaders first will probably make fixed function easier. So just learn shaders already!