I was having coffee in the graduate lounge when I overheard that “vibe coding” was Collins’ Word of the Year for 2025. 10 years ago, that term would have raised an eyebrow. Today, it felt ominously normal.

Not long after, Prof. Matthew Schwartz published what he called “Vibe Physics”: a record of supervising Claude through a real theoretical physics calculation from start to finish, without touching a file himself. Schwartz’s experiment was methodical, sustained, and exhausting in its own way: more than 110 drafts, tens of millions of tokens, over 40 hours of local CPU compute, and a final paper completed in two weeks. The striking part was not the physics, but the ethos of it. Schwartz positioned himself less as the primary researcher and more as an…..intellectual quality-control unit. Even then, his conclusion was not “AI has replaced the physicist”. It was that “taste” matters incredibly as frontier work can now be produced through prompting, supervision, and repeated correction. Domain expertise remains essential as there is, in his words, no going back. Autonomous scientific-agent systems such as CMBAgent have been demonstrated to carry out “PhD level cosmology tasks”, notably at the NeurIPS 2025 Fairverse competition, with authors explicitly framing them as vehicles for “autonomous scientific research”. Schwartz says LLMs could do graduate-level science circa 2027. This is naturally, delightful timing if you are set to finish your Master’s around then (I am). All this begs the question: Should scientists be buying more $200/month pro subscriptions and fewer ~$40k/year graduate students?

Answering these questions begs a (Hegelian) reification of humanity’s need to do astrophysics. In a recently published article, Prof. David Hogg asks with admirable directness: Why do astrophysics at all? Not what tools we create. Not how fast we can produce papers. Not whether the latest model can write a passable introduction section. Why do we do this weird, expensive, beautiful, occasionally colonial, occasionally militarized, very human thing called astrophysics?

Let me start, because how could I not, with Artemis.

When rockets rise, people stop. They look up. They drag their parents, their children, their partners, their neighbors into the moment. They gasp: on beaches, in living rooms, in front of screens with bad livestream. NASA’s Artemis II launched, sending four astronauts on the first crewed lunar flyby in more than 50 years; within days, the crew had flown farther from Earth than any humans before them. This matters not just because it is technologically impressive, but because it produces wonder at civilizational scale. We do not launch ourselves moonward just because it is cute. We do it because human beings seem utterly unable to leave the sky alone. Wonder matters. It is not decorative. It is not fluff. Wonder is one of the few things that unites us, briefly disarming cynicism and national borders at the same time.

Figure: Artist Sadequain’s Qalandar depicting humanity harnessing the workhorses of time to discover the universe.

But wonder alone is not enough. There is a case of why not to do astronomy. Astronomy has never been innocent. The same week an internship opportunity was being advertised at my school to work on ballistic missiles, Iranian universities, high schools, and homes were being bombarded. Astronomy does feed adjacent infrastructures, and those infrastructures are not always benign. Telescopes and satellites cultivate remote-sensing capacity. Precision optics do not ask whether they are serving wonder or war. Astrophysics cannot pretend to hide in its observatory dome, unstained by history. Hogg goes even further than I would by asserting astronomy to not have what he calls a “clinical value”: most of our measurements do not directly cure disease, feed the hungry, or house the displaced. Our observatories are not built on neutral ground either. It is easy to speak about the sublime while standing on someone else’s mountain. The modern history of astronomy includes mass displacement, extraction, exploitation, erasure, and enclosure. FAST, Maunakea, South Africa, Australia….and more. Astronomy compromises indigenous sovereignty, sacred land, and extends the long afterlife of colonial entitlement. If we are going to make a case for astrophysics, it has to be a grown-up case. No overglamorizing. Saganism, yes! but extended. No Sagan without politics. Wonder with memory. Awe with accountability.

It is not that astrophysics is justified because it is profitable, nor because it has some magical utilitarian output waiting around the corner. It is justified because people love astronomy. It creates human knowledge, universities should contain the whole range of serious human inquiry, it trains a technical workforce, it educates the public, physics itself needs astrophysics, and because doing astrophysics is a satisfying and meaningful human activity. But the most important point is the simplest one: people are always the ends, not merely the means.

When someone says an LLM can do something “better than a graduate student”, the phrasing is already flimsy. Better for whom? Better at what? Better if the point of a graduate student is to output polished artifacts as cheaply as possible? Hogg pushes hard against that language, and he is right to do so. Graduate students are not a temporary inconvenience on the road to efficient paper production. They are not artisanal GPUs with sleep schedules. The point of graduate education is not merely to “get the work done”.  Astrophysics is one of the purest examples we have of a field that people choose not because it is maximally remunerative, but because they want to do it. Hogg makes the wonderfully brutal point that if all we cared about were the answers, many of us would go make more money elsewhere and simply fund others to find the answers for us. The fact that we do not do this suggests something obviously profound: We are here not just because we want to know, but also because there is more to this tradition which includes the forging humanity’s next generation of thinkers. Tycho Brahe mentoring Kepler, Fowler Chandrasekhar, Bahcall Seager, and so many more.

That is the heart of the case for human astrophysics. We do not do astrophysics only to accumulate correct statements about the universe. We do it so the practice itself changes the practitioner.

This is why extreme positions on LLMs feel inadequate. “Let them cook”: models write the papers, run the code, draft the review, search the literature, maybe even propose the next program? Bah. The second extreme, “ban and punish” is just as unserious. We are not going to uninvent the tools. Nor should we. Refusing all AI use is not a sign of intellectual purity; often it is just a refusal to distinguish between assistance and abdication. 

So what is the middle way?

I keep returning to a line adapted from artist and writer Joanna Maciejewska: ‘’I want AI to do my laundry and dishes so that I can do [astrophysics], not AI to do [astrophysics] so I can do laundry and dishes’’. Although I believe a certain serendipity comes with laundry and dishes, so perhaps I would keep those as well.  In my own research, there are absolutely things I delegate to the scientific equivalent of laundry and dishes: fix this LaTeX error, make this plot prettier, explain this syntax, find where the bracket broke in this 679 line script. I am not above outsourcing drudgery. Frankly, I recommend it. If it can free up mental space for me to do higher level thinking, why not? The point was never asceticism but conscientious sovereignty. I am the one responsible for my work. I am the one responsible for the citation. I am the one who has to live with having become, or failing to become, an astrophysicist.

For me, the real question is not whether to use LLMs, but what a human must still learn in order to use them without becoming dependent on them. You need enough coding, statistics, physics, and writing ability to catch confident nonsense when it appears, and enough moral seriousness to remember that convenience is not innocence. AI is not free: it has substantial environmental and cognitive costs, and a way of making you feel productive whilst outsourcing your judgment. The task is not abstinence, but discipline in using these tools carefully, sparingly, and in ways that sharpen thought rather than replace it. A real middle way would ask not just what AI can do for astrophysics, but what kinds of astrophysicists its use is helping produce. If the next generation of astronomers cares about wonder, and if we are serious about military and colonial histories entangled with our field, then humanism cannot mean performative innocence; it has to mean drawing a line where we choose to work, what ends our tools serve, and how to use these tools with accountability, restraint, and a stubborn commitment to keeping people as the point.

People are not cost centers to be optimized. People are the project. Human astrophysics is not justified despite the fact that it is slow, social, difficult, and full of apprenticeship. It is justified because of those things. We need humane astrophysics because astrophysics is one of the places where humanity rehearses what kind of species it wants to be. One that looks up. One that collaborates. One that learns. One that remembers the cost of its instruments. One that refuses to treat human beings as expendable labor in a paper-production pipeline. One that can use machines without surrendering the beauty of thought.

Because it fills us with wonder.

Because it makes knowledge that belongs to everyone.

Because it trains people, who matter more than outputs.

Because if we surrender every intellectual task the moment a machine can imitate it, we will eventually forget what it was all for.

Because the universe is not just something to be computed. It is also something, still, to be encountered by people.

Because now more than ever, that distinction is worth defending.

Tell me why? Because…..I want it that way.

Editor Name: Annika Salmi

Cover and Image Credit: Syed Sadequain Ahmed Naqvi, The Sadequain Foundation.