Editorial Note
Recently entrepreneur Matt Shumer published “Something Big Is Happening,” a 5,000-word essay comparing the current AI moment to February 2020, right before COVID changed everything. It reached over 80 million people on X and was covered widely. Shumer wrote it for the people in his life who aren’t paying attention to AI, his family, his friends, the ones still asking “so what’s the deal with AI?” at dinner parties. His message was simple: wake up, start using these tools, because this is bigger than you think.
Shortly after, Bonnie Johnston at Blockbuster Blueprint wrote a response imagining a letter from a software engineer in 2029, someone who had done everything right, embraced AI early, became the “AI guy” at their company, and got replaced anyway. It’s worth reading.
Both pieces are aimed at people in technology and business. I’m a mathematician at a university. My world is different, the timelines are different, the fears are different, the questions are different. But I’ve been sitting with my own version of this anxiety for a while now, and reading those pieces made me want to write mine down.
What follows is written as a letter from my future self in 2029. Unlike the software engineer version, my narrator hasn’t arrived at clarity. The fears aren’t resolved. The questions are still open. That’s not a failure of imagination, it’s an honest reflection of where I actually am right now, trying to think clearly about what AI means for a career spent in harmonic analysis, operator theory, and several complex variables.
I don’t have good answers. But I think the questions are worth saying out loud.
A Letter From 2029
To the mathematician I was in Summer 2026
I’m not writing this from the far side of having figured things out. I want to be honest about that from the start, because the version of this letter that gets passed around tends to have a confident narrator, someone who was disrupted, rebuilt, and now has the clarity that comes with distance. I don’t have that. I’m writing this from inside the uncertainty, just three years further in, with more evidence and fewer answers than I expected.
Some of the fears you have right now have been confirmed. Some are still suspended. A few turned out to be wrong in ways that didn’t make things better, just differently hard. I’ll tell you what I can.
The fear you haven’t quite let yourself name yet.
There’s a specific one I want to start with, because I think it’s the one sitting closest to the surface for you right now, even if you haven’t said it aloud.
You’ve been working on certain problems for a long time. Years, in some cases. You know the terrain. You know where the difficulties live, which approaches have failed and why, what the obstructions feel like from the inside. And somewhere in the back of your mind is a question you don’t entirely want to ask: what if I missed something simple? What if there’s a move I didn’t see, and AI finds it in an afternoon, and I have to watch a problem I lived inside for a decade get dispatched by the right prompt?
I know this fear because I had it. I want to tell you it doesn’t happen that way. I can’t.
It happens sometimes. Not always, and not in the way your anxiety imagines, the AI doesn’t just announce the answer to your life’s work with cheerful efficiency. It’s more complicated and more disorienting than that. What actually happens is that the system finds an approach you hadn’t considered, which may or may not pan out, and you find yourself in the strange position of evaluating work that is adjacent to yours, that speaks to questions you own, that you cannot simply dismiss and cannot entirely claim. Whether that’s better or worse than being straightforwardly scooped, I honestly don’t know. It is different from anything I had language for.
What I can tell you is that the feeling of foolishness you’re anticipating, the retroactive embarrassment, the sense that the difficulty was illusory, that’s not quite what it feels like. The difficulty was real. The problems were hard. What changes is your relationship to hardness as a signal. For your whole career, the fact that something resisted you was informative, it meant the problem was deep, that the tools weren’t ready, that there was real mathematics in the gap. That signal becomes harder to read when the thing doing the resisting is you specifically, not the state of the field.
What happened to the research. The honest version.
I still work in harmonic analysis. The tools still matter, in some ways they matter more, because the applications to operator theory, function theory, and several complex variables have opened up faster than anyone predicted, and someone has to understand what the AI is doing when it reaches for a Carleson measure or invokes something from the theory of weights. That’s not nothing. But it is different from what I thought my career would be.
Here is what I found genuinely hard to anticipate: the pace doesn’t feel fast from the inside, and then suddenly it does.
In 2026, I was using AI for literature searches and occasional computation checks. It was useful. I was still the mathematician. By 2027, something had shifted. The systems could work with the mathematical literature at a depth that was alien, not just retrieving results but understanding how techniques transferred across subfields, identifying structural analogies between areas of analysis I’d worked in for years without seeing the connection. I had a collaborator suggest running a problem through one of the new systems, not to solve it, just to map the space. What came back was genuinely unsettling, a survey of adjacent results I hadn’t read, a candidate approach from a corner of function theory I barely knew, a decomposition that a brilliant postdoc might have suggested after a year in the area.
I kept working. I told myself that judgment, what’s worth pursuing, what constitutes a real result, the aesthetic of a proof, was mine. And for a while that was even true. But I want to say clearly what I didn’t understand then:
The skills hardest to articulate are not the same as the skills hardest to replicate. I had confused those two things for my entire career.
The intuition I’d built, for how operators behave on Hardy spaces, for what happens to harmonic measure near bad boundary points, for where the estimates in several complex variables were going to break down, that intuition was tacit because it was hard to transfer between humans. It turned out to be less hard to approximate from the accumulated literature, from the patterns in how the field had historically been surprised and corrected, when something could read everything and reason across it without fatigue. I believed that what I couldn’t easily explain was also what couldn’t be easily learned. That was wrong.
What I don’t know, still, in 2029, is whether the research I’m doing now means what research used to mean. I contribute. The work is real. But the question of what my specific understanding adds, beyond what the system would find without me, is a question I ask myself and don’t answer cleanly. I don’t know if that question has a good answer or if learning to live with it is just what mathematics is now.
What happened to teaching.
This isn’t the center of the story for me, research always was. But teaching mattered, and what happened to it was real.
The undergraduates changed first. By 2027, a student who wanted to understand a real analysis argument could have it explained, at exactly their level, with exactly the examples that matched their background, with infinite patience, by an AI that had modeled their understanding in real time. What I offered in a classroom was slower, coarser, and less adaptive. What I offered that was different was the experience of watching a human mathematician think, the false starts, the self-correction, the moments of genuine confusion that resolve into clarity. I believed that was pedagogically important.
I still believe it. The enrollment figures suggested the market disagreed.
The graduate students were more complicated and more painful. The best of them adapted faster than I did. They learned to use AI as a kind of mathematical ultrasonography, a way of imaging structure inside problems that would have taken months of manual work to map. Their papers got better. Their output was extraordinary.
What I found myself unable to give them was a clear picture of what their careers would look like. The academic job market, already brutal, had become something else, not because positions disappeared immediately, but because the implicit promise of an academic mathematical life was built on a theory of what mathematicians are for, and that theory was under pressure that hiring committees weren’t ready to discuss in interviews. I gave advice that was honest about the uncertainty and therefore not quite the advice they needed. I don’t know what the right advice would have been.
The questions I can’t answer for you.
You want to know if AI will solve the problems. I can’t tell you. Some of the problems I was working on in 2026 have been touched in ways I didn’t expect, not solved cleanly, but opened, reframed, connected to things that suggest the difficulty was in the formulation rather than the mathematics. Others remain exactly as hard as they were. I don’t have a principle that predicts which is which.
You want to know if the area will be exhausted. I would have said, in 2026, that mathematics is infinite and there will always be work. I said that confidently, the way you say things you haven’t examined. I’m less confident now. Not because mathematics is finite, it isn’t, but because the question isn’t whether problems exist, it’s whether the problems that exist will require what you specifically have spent your career building. The frontier moves. What constitutes interesting and tractable mathematics changes when the tools change this fast. Whether harmonic analysis as you practice it remains near the frontier, or becomes a well-mapped interior province, is something I watch and can’t predict.
You want to know if you’ll be passed by. I think this is the fear underneath the other fears, and I want to be honest with you: it’s possible. Not because you’re not good enough, but because the relationship between being good and being current is changing faster than any of us can adapt to, and adapting requires knowing what to adapt toward, and nobody knows that clearly. The people who are navigating this best are not the ones with the best answers. They’re the ones who have stayed genuinely curious rather than defensively certain, who have let themselves be interested in the new things even when the new things are uncomfortable. That’s not a guarantee. It’s just what I’ve observed.
The questions you should be asking right now that you aren’t yet.
Should you learn Lean? Probably yes, but I want to be careful about why. Not because formalization will protect your position, I’m not sure it will. Because the formal proof ecosystem is becoming the infrastructure through which mathematical knowledge gets organized and built upon, and not understanding it is like not understanding how papers work. Terence Tao thinks it transforms the field within five years. I think he’s roughly right on the timeline. What I can’t tell you is whether learning it will feel like gaining a powerful tool or like training your own replacement. Possibly both. Probably both.
How should you use AI so that it advances your research rather than replacing you? I have a partial answer, which is: use it most on the parts of the work you find least meaningful, and protect the parts you find most meaningful until you understand what you’re protecting. The literature searches, the computation checks, the verification of steps you already know how to do, those are fine to hand over. The initial encounter with a new problem, the period of genuine disorientation before you know what you’re looking at, the experience of being productively stuck, I’d be careful about compressing those with AI assistance too quickly. Not because the tools can’t help, but because some part of your understanding might depend on the duration of that experience in ways that aren’t obvious until the duration is gone. I’m not certain about this. But it’s the thing I most wish someone had said to me.
How should you advise the junior people working with you? This is the question that costs me the most sleep. I don’t have good advice for you, because I didn’t have good advice for them, and I’ve watched what happened when I gave advice calibrated to a world that was moving faster than I acknowledged. What I can say is: be honest with them about the uncertainty. Don’t give them the career guidance appropriate for the field as it was. Tell them to build toward problems they genuinely care about, not toward positions, because the positions are harder to predict than the problems. Tell them that the ability to ask good questions is going to matter more than the ability to execute answers, and that those are different skills and the second one is more trainable, which means the first one is more valuable. Tell them what you don’t know. They can feel it when you’re pretending otherwise, and the pretense makes the uncertainty lonelier rather than less.
The thing about meaning.
I almost left this out, because I don’t know how to say it without sounding like I’m being dramatic about something that is, in the end, a professional disruption rather than a tragedy.
But the meaning question is real, and I think it deserves to be named.
I used to have a particular experience doing mathematics, sitting with a hard problem, not knowing what I was looking at yet, feeling the shape of the difficulty without being able to articulate it. There was a texture to that, a companionship with not-knowing, that I didn’t have a name for because I didn’t need one. It was just what mathematics felt like from the inside.
I do mathematics differently now. I do more of it, in some measurable sense. I cover more ground. I engage with more problems and more connections between problems. And I find myself, sometimes, wondering if I’m doing the same thing I was doing before, or something that resembles it from the outside but is different in some way I don’t have the vocabulary to specify.
I’m not saying this to be melodramatic. I’m saying it because I think the meaning question is real and you should take it seriously, and taking it seriously means asking what it is about mathematics, specifically, in the texture of doing it, that matters to you, before the texture changes and you’re trying to remember what you wanted to preserve.
I didn’t ask that early enough. I was too busy being productive to ask what the productivity was for.
The mathematics being done in 2029 is extraordinary. Real things are being discovered. The connections between harmonic analysis and other areas are being mapped at a scale that would have seemed impossible in 2026. I find it genuinely beautiful, which makes everything I’ve written here stranger to hold.
I don’t have a clean ending. I don’t have the advice that will make this navigable. What I have is this: the uncertainty you feel right now is not a failure of nerve. It’s an accurate read of the situation. The people who were most confident in 2026 about what would and wouldn’t change were, in my observation, not better prepared, they were just more surprised when they turned out to be wrong.
Stay uncertain. Stay curious. Ask the questions you’re already asking, and ask them out loud to the people working with you.
That’s not enough. But it’s what I’ve got.