Unless you are working on ChatGPT-6, i assure you, you can definitely estimate how long things will take. your estimates will not be perfect, but they will be good approximations for guiding development.
Even OpenAI by now probably is pretty good at estimating how long the next ChatGPT version will take to develop, since they've done it 4-5 times already.
Alway always distrust someone who says "we are doing things that have never been done before". At the very bleeding edge (OpenAI, SpaceX), you are iterating on something that has been done many times before, but with a small innovation.