as optimised as possible
The key element of any optimisation is defining the cost function in the first place. You can’t just say “I want it more optimised”, that’s completely open-ended. Or rather, you can, but there’s nothing further that can be done. It’s like saying “I want things to be better”. Don’t we all?
What is a “good” layout for this function is really hard to even define on the first place. Total ratsnest length is a quite naive metric with many modern boards, as if you have power/ground nets, much of that length will vanish once you drop your vias. Some other signals don’t care a jot about their length, for some it’s critical. And so on though the many many other constraints a PCB embodies - component sides, heights, weights, heatsinking, crosstalk/EMI, layer counts, via sizes, ability to sneak though rows of pads (DIPs: yes, QFNs: no), etc., etc.
I also don’t personally find the autoplace function useful, and it doesn’t seem (again, to me personally) to behave in a well-defined enough way that I feel able to reason about it as part of a wider workflow. Whereas P&M is well-defined - you know what will happen before you press the button, and the fairly orderly arrangement is easy to grab bits from. I suspect auto-place dates from when boards were more often a bunch of spidery traces directly connecting big THT parts on a 1/2 layer board rather than the modern multi-layer plane-centric style full of tiny SMD parts.
Coming up with useful, by which I mean not only is the result “good”, but the UI to get to it is also fluent, alternatives, however, is very hard, even for quite limited targeted sub-tools. I suspect it’s impossible to do it all at once universally. Perhaps the AI crowd surprise everyone, I suppose, but their results so far still look like a confused 90s auto-router, with fancier websites.
All that said, if you have a concrete idea for how the tool could be better, a clear description of how you think it could work in various practical scenarios, what scenarios aren’t covered, how the UI works (one tool, many separate tools, dialogs, wizards, etc) and so on, that actually makes it possible to think about it meaningfully. “Couldn’t it be better?” isn’t really that.