Clearly if we have two operators $\phi(t_1)$ and $\psi(t_2)$ and define a time ordering operator $T$ acting on operators such that $$T(\phi(t_1)\psi(t_2)):=\phi(t_1)\psi(t_2),~\text{if $t_1>t_2$ otherwise the other way around}$$ we could pretend that inside $T$ these operators commute. However I am not quite convinced that we are allowed to simplify expressions inside, as there might well be such operators $A$ and $B$ that $A(x)B(y)$ is non-zero for all $x,y$ but $B(y)A(x)$ is always zero. Thus if we choose to simplify expressions inside $T$ it may well go wrong as we end up with a null operator, and $T(0)$ certainly will not return you with a time-ordered non-zero result.
What is going on here?