The premise is seductive. AI tutors don't get tired. They don't have bad days. They adapt in real time to what a student understands and what they don't. They scale infinitely. They cost a fraction of what a human math teacher costs. For investors betting that education technology is the next trillion-dollar market, it's perfect.

But here's what algorithms can't do: they can't watch a student's face light up when they suddenly understand why the Pythagorean theorem works. They can't sit with confusion long enough for genuine insight to emerge. They can't tell the difference between a student who's memorizing steps and a student who's thinking.

At Business Insider's The Long Play event, Hong explained how she has attracted top talent from across Silicon Valley—engineers, mathematicians, learning scientists. The team is formidable. The vision is grand. And yet.

Mathematics, at its core, is a language for thinking. It's a way of describing relationships, patterns, structures. The best math teachers—the ones students remember decades later—aren't the ones who explained the most steps clearly. They're the ones who asked the right questions, who let students stumble toward understanding, who made it feel like discovery rather than instruction.

This is precisely where the Silicon Valley model breaks down. Education Week reports that as AI tools are integrated into math education, teachers and students will need to develop a healthy amount of skepticism about the limitations of AI tools. The publication's phrasing is diplomatic. What educators actually say is sharper: skepticism isn't enough. Critical resistance is the baseline.

The worry isn't that AI is bad at math. AI is excellent at math—better than most humans, probably. The worry is that the presence of a machine that can instantly provide the right answer changes what students learn to value. Effort becomes pointless. Struggle becomes something to avoid rather than work through. The mathematical thinking—the actual muscle of mathematics—atrophies.

There's also the unsettling reality of what gets lost in scale. Human teachers are inefficient. They spend time on tangents. They follow student curiosity down rabbit holes that have nothing to do with the curriculum. They adapt on the fly. A teacher might spend 45 minutes on one concept because three students needed her to approach it from a different angle. An AI system optimizes that away. Faster. Better. More efficient. Less human.

Hong's ability to attract world-class talent suggests the engineering problems are solvable. The billion-dollar valuation suggests investors believe the market is massive. Neither of these things proves the educational premise is sound.

"Teachers and students will need to develop a healthy amount of skepticism about the limitations of AI tools." — Education Week

The equity argument is the most seductive selling point. If AI tutors could provide personalized, patient instruction to every student—regardless of where they live or how much money their family has—that would be genuinely transformative. The problem is that this vision assumes the problem with math education is access to instruction. It's not. The problem is that we've turned mathematics into a series of procedures to memorize rather than patterns to understand. Throwing AI at that problem just accelerates the damage.

None of this means Axiom Math will fail commercially. Venture capital doesn't reward pedagogical purity. It rewards growth, adoption, network effects. A tool that gets kids to pass standardized tests faster will win the market, regardless of whether it's building mathematical thinking.

What it does mean is that we're about to run a large-scale experiment on an entire generation of math students. The control group won't exist. We won't know what was lost until we notice what's missing: the students who don't think mathematically, who see numbers as alien symbols to be decoded rather than tools to think with, who got very good at following rules and terrible at breaking them creatively.

Mathematics needs humans in the loop. Not as instructors to be replaced, but as witnesses to the messy, inefficient, irreplaceable process of actually learning something. The algorithm can't be the teacher. It can't be the classroom. It can't be the subject itself.

If Axiom Math becomes dominant, we'll have built something very efficient. Whether we've built something educational is a much harder question—and one that won't be answered until it's too late to ask it a different way.