I mean, I question if there could species that chose from their own gens, those who had been more useful during their life and increasing the chance that their descendants had them since their conception.
There are two kinds of argument against this - one from basic principles, and one from empirical evidence.
As far as basic principles, I think Dawkins makes very good arguments against Lamarckian evolution in his book "The Selfish Gene". Essentially he points out the example of muscles getting larger when we exercise, and notes that not only isn't this universal - plenty of structures get worn down with use, not strengthened - the fact that muscles specifically do this is itself an adaptation. And he points out there is no mechanism for this adaptation arising through Lamarkian evolution if it isn't already there. There's a basic bootstrapping problem involved that Darwinian evolution doesn't have.
When we consider your example the same issue arises. By what mechanism would an organism choose which genes to pass on? Can you? Of course the fact we don't have conscious control over the genes we pass on isn't necessarily dispositive, our bodies do plenty of things we don't have conscious control over. But not only would there need to be a mechanism by which some genes are selected to pass on and not others, you would need some way for the information about which genes to select to get back to said mechanism. Making this choice is an incredibly complex issue... The things you'd have to know about the environment, the ways genes impact the final phenotype, which phenotypes are truly optimal for the environment... We humans aren't at the point of doing this. Natural selection works out for extremely dumb reasons. You are proposing a system that does this in a smart way... We already know how evolution comes up with smart systems, that integrate information about the environment and infer general rules about how the world works and predict the best actions in consequence: it makes brains. The brains it's made so far that are best at this are ours, and "explicitly choosing which genes to pass on" isn't what we're using them for so far. That might come in future decades with genetic engineering, and when it does it will be an entirely novel form of evolution. It's incredibly implausible to suppose that other organisms have been doing this all along.
Now, could some genes or gene families happen to have some kind of positive feedback property that means they get selected more when they're useful and selected against when they're not, more than natural selection could predict? Maybe. It's not like organisms that use "meta-properties" of evolution, like tuning their mutation rates depending on the environment, don't exist (see this abstract on bacteria for example: https://pubmed.ncbi.nlm.nih.gov/16677295/). That could too. But it would apply to a few specific genes in those organisms, it wouldn't be a general evolutionary property.
As for empirical evidence, the thing is the "neutral hypothesis" can be tested, because random behavior has (paradoxically) very predictable properties. I think we can generally say that in everything we see about mutations appearing, the inferred mutation history seen in DNA patterns, in the evolutionary history of various phenotypic traits, that all those things point to mutations happening randomly with respect to usefulness. But one specific example I like very much is the Lenski experiments:
http://myxo.css.msu.edu/ecoli/
What happened is that Lenski ran a long-term experiment looking at the evolution of certain bacteria, by culturing them in a certain environment and, and this is key, freezing batches every few generations to have a replayable "record" of the state of the bacteria at various points. The most famous aspect of this experiment is how some of the bacteria evolved the ability to break down citrate, a completely novel and even definition-breaking ability for this type of bacteria. And here's the interesting aspect that's relevant to our discussion here: when they looked back at previous frozen generations ancestral to the specific bacteria that did this to figure out what happened, they found that if they cultured bacteria from before a certain point, almost none would develop the ability to break down citrate. But if they cultured bacteria from after that point, plenty would. Lenski deduced that the ability to break down citrate required several mutations, and one of those mutations was developed in the ancestor at that point. Hence, all the ancestors before that point were N mutations away from developing the ability to break down citrate, but those after it were only N-1 mutations away, and so they developed it much more readily.
Now how is this relevant? This difference between before and after the potentiating mutation would only happen if the mutations were occurring randomly, as opposed to "with the goal of developing the ability to break down citrate". If the bacteria were trying to break down citrate from generation 0, and acquired those mutations for that specific goal, then there would be no difference in which of the ancestors you looked at - they'd all be on the same path, and all equally likely to reach the goal. It wouldn't matter that the older bacteria were one extra mutation away from the adaptation than the later ones: they were planning on getting that mutation anyway. It's like, say you have a toddler and you want them to walk to a chair, but they don't control their walking very well; they're weaving all over the place, it's frankly not clear they even want to get to the chair to begin with. Where you set them down will make a big difference in whether they reach the chair or not: if you put them close to the chair, their random movements are quite likely to take them there, but the further away they are the more likely they are to just wander off wherever instead. Now say instead you have a hungry 6 year-old, and you put their favorite candy on the chair where they can see it. That kid is making a beeline for the chair, and it doesn't really matter where you set them down - if they're far away it might take them more time to get there, but not much, and they are as likely to get there in the end as if they'd been set closer.
The evolution of citrate metabolism in Lenski's bacteria behaved like the aimless toddler, not the focused 6 year-old.
ETA: Actually I kind of want to expand, add yet another element to the answer, which is: why this might not even be desirable. I see other comments have given examples of genes being transmitted non-randomly. I myself didn't exclude genes being transmitted non-randomly with respect to usefulness, but said it would be a property of specific genes in specific organisms, not a general feature of evolution, and I stand by that. I'll add that I don't expect this kind of property to be an important part of developing novel adaptations in particular, and that I expect big caveats on how smart these features are. I don't know what level of intelligence OP is considering, but I'd guess they're thinking, "wouldn't evolution work better if it chose adaptations more intelligently, like we humans would, instead of randomly?", mostly because it's what I myself would think. And that's the idea I want to challenge here.
I said earlier that natural selection works out for extremely dumb reasons. That's a weird thing to say, so let me explain: natural selection works because it's constantly testing things against reality. By definition; "natural" selection is "selection by nature". On average, things get passed on if they actually help, exactly insofar as they actually help. This is what makes evolution by natural selection so powerful, and come up with such innovative solutions that would never occur to humans: everything in reality is on the table (or at least, limited by the fitness landscape, not cognitive biases), and only things that work in reality are kept.
But that power is also a limitation. Because how is it that humans design that's different? We also test our designs against reality: we build them, check that they work, and keep them if they do. But "trial and error" is notoriously slow and inefficient. The power of our design process, the intelligence of it, is that we don't just test our designs against reality: we test them against imagined realities. We imagine how the world could be, we imagine different ways designs could interact with an abstracted mental model of how the world is. This is what allows us foresight and planning... which is something evolution by natural selection does NOT have, and which is why we can do things it can't. And I think that's the kind of intelligence people who ask "what if evolution was intelligent" are thinking of. But it's also a danger, because it means we can go full forward with an idea, and find out too late that it had awful consequences we didn't anticipate. It also bumps against the limits of our own imagination, which will never encompass the whole of reality. This is the big fear with genetic engineering, and it's why I'm not convinced that selecting genes for adaptivity would be as useful to an evolving organism as just blind natural selection would. You'd gain some things (speed, ability to cross fitness valleys) but lose others (reliability, ability to find fitness paths the foresighting system did not foresee). We might pull it off, but that gets to what I was saying earlier - we, as a species, are the smartest, best-at-planning-and-forecasting system nature has come up with yet. If we're barely, debatably at the point of being able to do this, what existing biological system could do better?
