Optimal representative participation in the age of AI/ML?

Just answered the great questions of the 2022 Stack Overflow Developer Survey and earned the Census - Badge, here I see the number of filled surveys and as an enthusiastic beginner in data science I am curious about the optimal representative participation in case of 17,866,773 total users!?

I did a quick research, but stuck at the following sentence:

Once the population exceeds 20,000, your sample size will not change very much anymore.

Please help me to understand this (agree/disagree welcome to) and revela your experience with more than 20,000 samples and do these experiences are true in the age of AI/ML supported data processing?