Agreed; most programmers I know who write a lot of x86 Asm tend to stick with the Intel syntax. I've never found the "it's easier to parse" argument for AT&T syntax particularly convincing, especially as GAS is written in C while some of the Intel syntax assemblers existing at the time were themselves written in x86 Asm. Neither is the "it makes it more consistent with the other arch's assemblers" - just look at GAS' syntax for MIPS and ARM. It seems to me like someone was really obsessed with the SPARC syntax (which I find roughly as horrendous.)
I can agree with most of that, except for the fact that Intel arguments are backwards. That throws me off every time. I don't know of any other assembler syntax that uses that argument order.
For me, I remember the notation by correlating it to its "high-level" equivalent.
mov eax, [ ptr ]
is like
eax = *ptr; // or, eax = ptr[ 0 ];
The offset/multiplier memory addressing format for AT&T syntax was always more troubling for me. Coming from a TASM/MASM/NASM/PASCAL/x86 background first, it felt "icky" to put offsets outside of the "brackets" (or parenthesis, as it were) [0][1].
Standard ARM, MIPS, PowerPC, x86, and Z80 Asm syntax all have the destination on the far left.
68k, Alpha, PDP-11, SPARC, and VAX have the destination on the far right.
The order is probably as contentious as the great endianness debate, but I think one of the most awkward parts of having src, dst order is that subtraction looks backwards. I prefer dst, src because it corresponds closely with the direction of assignment in higher-level languages:
http://x86asm.net/articles/what-i-dislike-about-gas/