I’ve gone to almost every AI-adjacent event I could find on campus this semester. I immersed myself in podcasts and scrolled on X. Some talks were not centered around AI, but guest speakers often couldn’t avoid talking about it. In between were countless conversations with friends — founders and wannabes, bankers-to-be, researchers-to-be, and classmates whose worldviews I had to slowly come to understand. By the end of April, I had my notes app full of scattered thoughts and two things I could not stop turning over.

One. We have to reimagine the career structures we are used to.

In his book “Zero to One,” Peter Thiel observes that we have learned to hoard options instead of making them — hedging against an unknowable future by becoming broadly competent at nothing in particular, ready for any well-trodden path. I read it in January and recognized myself. The strategy worked because the destination was assumed to be there: a long, predictable career ladder that would pick you up once you had stocked enough credentials at the bottom. You could afford to be undirected at 20 because the structure would direct you at 25.

The well-trodden paths used to be safe. Now AI is disrupting them. Companies are losing the economic incentives to hire junior workers at all — when a model can do the research memo, the first-pass code review, the entry-level analyst work. What disappears with those jobs is the old apprenticeship structure. Entry-level work was never just where the young did the cheap labor while the senior workers did the thinking. It was where young people became senior. You would watch a partner reframe a problem in a meeting and feel something click. You would learn by observing the kind of judgment you cannot acquire from reading a synthesis. It had to be lived. The apprenticeship structure was the staircase where expertise climbed across generations.

This means we cannot just stock credentials and wait for the structure to direct us, because the structure is no longer going to be there. The question left for us, the young generation, is, “How does a 22-year-old new grad gain strategic judgment without the years of dirty-work practice that used to build it? How does a 22-year-old get a foot inside an industry when firms are no longer incentivized to hire them at the bottom?” This is a problem of how a society reproduces its own expertise across generations — and we must now invent an answer.

The good news is that the same disruption hollowing out the bottom is also flattening the top of the career ladder. In a Sequoia article, Jack Dorsey and Roelof Botha advocate for redesigning companies as flat networks of small teams supported by AI rather than tall hierarchies of middle managers. A flatter structure means there may be more room for young people to contribute earlier, be heard sooner, and learn faster than the apprenticeship model ever allowed — if we figure out how to step into it without the old training that used to get us there.

Two. There are alternatives, and we will need imagination to find them.

One of my favorite classes this semester was a history class on AI by Professor Uluğ Kuzuoğlu. It walked us through all the paths contemporary AI did not take: expert systems, neural networks at moments when they were almost abandoned, Japan’s fifth-generation computer system that was once expected to define the future but stalled. The technology we now call AI — large language models, transformers, scaling — is not the only AI history could have produced. It is one paradigm that won, among other possibilities, in a particular decade, under particular incentives, and with particular sources of capital. And large language models are already only one piece of the wider field. World models, embodied AI, alternative architectures we haven’t named yet… the future is more open than just one linear path.

The same narrowing is happening again, in real time. We are watching frontier companies race toward AGI — artificial general intelligence — and claim their mission is to benefit all humanity. But OpenAI’s own charter defines AGI as the AI that can outperform humans at “most economically valuable work.” Not all valuable and intelligent work is economically beneficial, and we must be wary of this reduction of intelligence to monetizability.

There are other ways to build and other things to build for. One alternative comes from journalist Karen Hao. When she came to WashU last fall, she made the case for task-specific AI systems: narrower models built to solve particular problems rather than general-purpose ones built to capture markets. Her example was AlphaFold, which won the Nobel Prize in chemistry in 2024 for predicting protein folding — smaller, more sustainable, and more useful than a general-purpose model. AI does not have to be a winner-take-all race for general intelligence. It can be a set of careful tools for specific human problems.

If, as college students, we want to stand at the edge of the future, we cannot just drift along with every new model coming from Anthropic or OpenAI. We have to imagine alternatives — what other AI futures are technically possible? And what new structures could replace the old ones? The answers are not being passed down. We have to imagine them ourselves.

Where I’m landing

So much more needs to be reimagined than what can fit into this article: wealth distribution, labor markets, and how we live our lives. Thanks to the talks that echoed or provoked my other reimaginations this semester: Nicholas Thompson, Ocean Vuong, Phil Gramm (I still doubt his claim that AI lifts everyone equally), and many more.

Over the course of this semester, I absorbed as much as I could, but I still do not know what will happen. I take some comfort, though, in the fact that in any given week on campus, someone is speaking and someone is listening. This is a small structure, very old, that doesn’t need reimagining. We listen, we let it complicate us, and we try to write down what we have heard. If everything else must be reimagined, this is where we begin.