I tried the method suggested in the comments, but it seems to bog down.
$$\begin{align}
I&=\prod_{k=1}^{89}\cos{k^\circ}\\
&=\cos{45^\circ}\prod_{k=1}^{44}\cos{k^\circ}\prod_{k=46}^{89}\cos{k^\circ}\\
&=\frac1{\sqrt2}\prod_{k=1}^{44}\cos{k^\circ}\cos(90-k)^\circ\\
&=\frac1{\sqrt2}\prod_{k=1}^{44}\frac12(\cos90^\circ+\cos(90-2k)^\circ)\\
&=\frac1{2^{44}\sqrt2}\prod_{k=1}^{44}\cos(90-2k)^\circ\\
&=\frac1{2^{44}\sqrt2}\prod_{k=1}^{22}\cos(90-2k)^\circ\prod_{k=23}^{44}\cos(90-2k)^\circ\\
&=\frac1{2^{44}\sqrt2}\prod_{k=1}^{22}\frac12(\cos90^\circ+\cos(90-4k)^\circ)\\
&=\frac1{2^{66}\sqrt2}\prod_{k=1}^{22}\cos(90-4k)^\circ
\end{align}$$
and now we can no longer group the arguments in pairs that add up to $90^\circ.$ At the next stage, if I follow the strategy of grouping the largest arguments with the smallest, I get $$
\frac1{2^{77}\sqrt2}\prod_{k=1}^{11}(\cos88^\circ+\cos(92-8k)^\circ),$$ which is better than what we started with, but still unpleasant.
I think it might be better to approximate the product as follows.
$$\begin{align}
\log I&=\sum_{k=0}^{90}\log(\cos k^\circ)\\
&\approx \int_0^{90}\log\cos\left(\frac{\pi x}{180}\right)\mathrm{d}x\\
&=\frac{180}{\pi}\int_0^{\pi/2}\log\cos u\mathrm{d}u
\end{align}$$
It is known that $$\int_0^{\pi/2}\log\cos u\mathrm{d}u=-\frac{\pi\log2}{2},$$ so we get $$\log I \approx -90\log2$$ and $$I\approx2^{-90}.$$
One can use the known error estimate in the trapezoidal rule to bound the error in this estimate, but I haven't done that.
NOTES
This may not be a very good estimate, at least in terms of relative error. A simple-minded python script gives $$I\approx 2^{-85.75}.$$ Also, if we're going to approximate it, we could just say $\cos88^\circ\approx0$ and forge ahead.
P.S.
Whoops. The derivative of $\log\cos x$ is $-\tan x$, so bounding the error with the trapezoidal rule, or the Euler-Maclaurin formula won't be easy.