People often say things like “you won’t be replaced by AI, you’ll be replaced by someone using AI”. I admit it’s a snappy line but I think it is a dangerous oversimplification of what’s happening in the near future.

If current trends in AI and knowledge work hold, then agentic systems will be capable of automating a very large swath of tasks of basically all non-manual labor in the next several years.

The current trend, though, also shows us that the machine is lacking in taste and creativity.1 It can perform superhuman feats of mathematical reasoning, but not invent an entirely new field of mathematics. It can, in my experience, write technically competent philosophy papers that are mild extensions of existing ideas, even in novel and creative ways, but not yet anything that could wake one from their dogmatic slumber. The machine is capable of following ideas to logical conclusions faster and more consistently than us, but the lateral thinking needed to find a good rabbit hole to dive down seems to elude it. As of now in September of 2026, when I’m writing the post you’re reading, this is main reason why AI labs are claiming to have automated research assistants but not automated researchers.

So if we grant that the AI can’t yet perform acts of true novelty, although admittedly such field-creating world-shaping acts of creative novelty are rare to begin with, this has some obvious implications for how we’ll do knowledge work.

One is that more of our work in general will become a kind of research, with higher level investigation, abstraction, and experimentation. Implementations of ideas become cheap, modulo bureaucratic slowdown2, and you’re slowed down more by your ability to imagine things worth trying than in the act of trying them. This means our valuable time will be spent in deeper work and study.

In order to evaulate the work the machine can do, though, we cannot surrender our committment to domain expertise. I disagree with anxiety driven predictions of people like Cory Doctorow, who think our relationship with the machine will inherently devolve into a “reverse centaur” where we start reflexively approving its work without being able to understand or contextualize it, simply because that’s a path of least resistance. I see this as a deeply cynical view of people and, to me, the only reasons why this “I push the button” relationship to the machine would happen are larger systemic pressures to work and publish and deploy as fast as possible. I do not think these are a given and, honestly, I think it would only happen in the transitionary period as we figure out the healthier relationship to our strange little coworkers: one of wild ambition, experimentation, and effectiveness.

What does this mean for the future knowledge-worker? The employable person will be spending more of their work time

  • reading
  • synthesizing
  • brainstorming
  • explaining

than ever before.

So rather than saying that I think you’ll be replaced “by someone using AI” I think it’s morely likely to be “you’ll be replaced by someone who is a better, more novel, thinker capable of deep understanding of the subject domain and yet broad enough in their background to connect non-obvious ideas to the problem at hand”.

That’s not very catchy but I think it’s closer to the truth!

A consequence for us educators is that we need to focus on giving students the opportunity to do more “research shaped” work: more reading, more synthesizing ideas, more creative projects, more opportunities to swing big and potentially fail in a constructive way.

A consequence for students, oh poor students, is that you cannot coast. The degree by itself doesn’t help you, it doesn’t necessarily signify anything. You’ll need to demonstrate a capacity for critical thought and expertise directly; but fortunately the process of getting the degree gives you a space to explore and build up all of those skills. When you’re given avenues to explore that are optional in your classes, try them! Take your professors up on those opportunities!

I argue that we will all need to be, in the literal sense, philosophers: lovers of knowledge! I also think in twenty years there will be students entering my classroom who were raised in a system that has already reckoned with these changes and the strangeness of this technological revolution; but, for now, this will be a hard transition and I’m sorry for that—but it will also be an exciting time as we’re all trying to figure out what comes now!

So to my fellow educators and faculty what do I suggest? We must learn to help ourselves, to be stranger and more ambitious thinkers: spend more time on side-projects, keep a proper notebook of ideas you’d have never tried before, try to learn new fields and expand horizons. We do that work and we take the fruits of it into our classrooms.

To students I say read. Read as much as you can on everything. Be voracious for every intriguing thing that passes your way. I want you to find out what the open problems and wild ideas are in your field(s) of study, to start thinking of how you’d tackle them. I need you to try to understand as much as possible as deeply as possible, to treat your academic time as not just an opportunity to get a credential but like the rare chance to drink your fill of knowledge, so that you’re ready to do the previously unimagined.

  1. That may change! I cannot emphasize enough that it’s very hard to predict where this is going, and it might be a losing bet to argue that there’s anything the deep learning revolution won’t master within a decade. That, however, means we would have reached actual super-intelligence and, honestly, who even knows what happens if we hit that point. It’s clear that there’s internal pressure within AI research to stop short of super-intelligence until we’re certain we can handle it without it doing something weird. ↩

  2. I’m leaving for another post to talk about how I think Stafford Beer’s theories of Cybernetic Management are about to have a new form, helping us manage the deluge of information we are now capable of creating ↩