I always struggled with the automation question:
Should I just do X in 5 mins or should I automate it (with code, a system etc) in 60 mins so it takes 1 min in the future?
After doing a PhD and running a startup I think the answer is
"almost always automate".
Details

Forget solving this like an equation. It never works. The theory goes:
If doing it N times, automate if
( 5 - 1 )*N > ( 60 - 5 )
ie time saved > time spent

But you don't know N (you'll tend to underestimate it). You'll also be wrong about the other estimates for that matter (you'll understate those too)

The big thing though is that you'll underestimate the savings from automation. For a few reasons:

1) Manual tasks get faster but are also error prone which is a linear multiplier of lost efficiency

2) You are probably underestimating how much of it can be automttd. You may automt a part, then you find a new bit to be automtd, then you hook automtns up to each other then you discover a principle to automt things in the future and this is an exponential increase in efficiency

3) Automation allows you to forget about them and think about new stuff. This is also an exponential benefit

When don't you automate? You should never blanket refuse to automate, but there are rare cases I've noticed it best to wait
1) If it would take > a week it's generally a complicated automation that's risks losing its advantage (but you can break it into small automations

2) If you don't know how to automate it, you don't know enough about it. Keep "doing things that don't scale" in that case until the path is clear