June 14, 2026
Proof or it didn’t happen
Formal Methods and the Future of Programming
Wall Street coding giant flips the script, and the comments are not buying the hype
TLDR: Jane Street says AI coding tools have made it worth revisiting super-rigorous software checking, and it’s now building a team around that idea. The comments were split between curious optimism and blunt skepticism, with many asking whether this really prevents mistakes or just moves the mess somewhere else.
Jane Street just did a full-on plot twist: after roughly 25 years of saying advanced proof-based software checks were too expensive and too impractical, the firm now says the rise of AI coding assistants has changed everything. In plain English, the company thinks if machines are going to write more code, humans may need better ways to double-check that code before it causes chaos. Cue the internet instantly turning this from a strategy update into a spicy town square debate.
The biggest split in the comments? One camp is intrigued, but wants receipts. How exactly does this help people move faster with AI-written code? commenter eddiepete basically asks whether we’re just replacing one messy pile of code with another messy pile of "verification" work. That skepticism hit hard. Another recurring mood was: nice idea, but please stop acting like this is magic. xvilka pointed to the famously verified seL4 project still having nasty bugs, which became the thread’s unofficial reality check meme: even the ultra-careful stuff can still go sideways.
Meanwhile, others saw a bigger cultural shift. jdw64 boiled it down to a dramatic labor rebrand: if AI writes more of the software, then human value moves to checking it. That sparked a more philosophical vibe about what programming even is now, with a side note about how hard all this already is for non-English speakers. And in classic comment-section fashion, someone casually dropped a link claiming “formal methods won, now what?” — the kind of drive-by hot take that practically begs for a sequel.
Key Points
- •Jane Street says it has changed its long-standing view and is now building a team focused on formal methods.
- •The article says Jane Street historically viewed full formal methods as too costly for most software, despite strong support for tools such as type systems.
- •Yaron Minsky cites the seL4 microkernel as an example of formal verification’s value and expense, noting 25 person-years were spent verifying 8,700 lines of C.
- •The article argues that agentic coding lowers the cost of using formal methods by helping more people use these tools productively.
- •Jane Street says verification is becoming more important as models get better at generating useful code, increasing the potential benefit of formal methods.