Yaron Minsky spent a quarter century telling anyone who asked that Jane Street had little interest in formal methods. The firm built its trading systems on OCaml and sophisticated type systems instead. Those tools delivered speed, reliability and control without the crushing overhead of full proofs. Times change. On June 7, 2026, Minsky published a blog post reversing that long-held stance. Jane Street is now assembling a dedicated formal methods team in New York and London.
The announcement comes at a moment when quantitative trading firms face mounting complexity. Markets move faster. Systems grow more interconnected. And large language models have begun injecting code at a pace human reviewers cannot match. The old calculus no longer holds.
“I’ve been telling people for the last 25 years that Jane Street as an organization was just not interested in formal methods,” Minsky wrote in the Jane Street Blog post. “I’m not saying that anymore.” He acknowledges the firm was not wrong before. The costs simply outweighed the gains for most trading infrastructure. The seL4 verified microkernel stands as exhibit A. Researchers invested 25 person-years to prove 8,700 lines of C. Each line demanded roughly 23 lines of proof and half a person-day of effort, according to the original SOSP 2009 paper.
Such expense made sense for a security-critical kernel. It did not for the fast-moving, constantly evolving codebase that powers Jane Street’s trading, risk and execution systems. Type systems provided enough guardrails. Property-based testing, fuzzing and a culture of expect tests filled the gaps. The firm even open-sourced tools to make those tests more effective.
Agentic programming upended the equation. Models now generate functional code at scale. They do so quickly. Yet the output often arrives bloated, littered with edge cases and blind to the codebase’s deeper invariants. Human engineers spend increasing time reviewing, refactoring and verifying. That verification bottleneck threatens to erase much of the productivity gain AI promises.
Formal methods offer a different kind of feedback. They deliver universal guarantees rather than sampled evidence. A well-designed type system can eliminate entire classes of bugs. Prevent data races completely. Make cross-site scripting impossible by construction. Agents respond well to this kind of crisp signal. During reinforcement learning and interactive coding sessions alike, precise feedback improves performance. Tests help, but they cannot cover every state. Proofs can.
“A lot of why we’re excited about full-on formal methods is that we see how valuable types are when programming with agents, both for easing the verification bottleneck and providing agents with better feedback,” Minsky explained. The firm’s new team will pursue that vision aggressively. The goal is to make formal methods as routine a part of the developer experience as advanced types are today.
Jane Street posted openings for formal methods engineers shortly after the blog appeared. The New York role, listed on the company’s careers page, seeks candidates with deep experience in interactive proof assistants, automated theorem provers, static analysis, refinement types or program logics. Top software engineering skills matter more than prior OCaml knowledge. The team will integrate existing tools, adapt the firm’s OxCaml fork and explore novel language extensions that embed specifications and proof techniques directly into the type system.
Control over the language gives Jane Street an edge few organizations possess. Most companies consume programming languages as fixed products. Jane Street maintains its own compiler fork and evolves it in tandem with its needs. That flexibility lets engineers experiment with ownership disciplines, modular specifications and tighter coupling between code and proof. The firm’s developer culture supports the shift. Engineers already clamor for stronger type features. They possess the mathematical maturity to adopt proof tools without widespread resistance.
The move also reflects broader industry undercurrents. Other financial firms have quietly increased investment in program verification for compliance, model risk and operational resilience. Yet Jane Street’s public reversal carries particular weight. The firm has long stood as a champion of functional programming in production finance. Its embrace of formal methods could accelerate adoption elsewhere. Recent discussions on X highlight the excitement. One post noted that “the AI writing your buggy code is what makes formal methods viable.” Another simply shared the blog with visible enthusiasm from the programming community.
Ron Minsky, who has spoken about Jane Street’s AI practices at Bug Bash 2026, reinforced the theme in related talks. Traditional engineering disciplines have grown more vital, not less, in the presence of generative models. Types, tests and simulations remain foundational. Formal methods extend that foundation into territory agents can exploit more effectively.
Challenges remain. Proofs still require human guidance for high-level strategy. Models excel at filling in details but falter on complex invariants without direction. The new team must bridge that gap. It must deliver tools that practicing engineers actually use rather than academic exercises. Early efforts will likely focus on high-impact areas such as core trading libraries, risk calculations and infrastructure that agents modify frequently.
Jane Street has a track record of turning research ideas into daily practice. It introduced OCaml to production trading two decades ago when few considered functional languages suitable for finance. The firm later open-sourced large parts of its infrastructure and contributed to the broader OCaml ecosystem. That pattern suggests the formal methods initiative will emphasize practicality. Success will be measured by fewer production incidents, faster safe iteration and measurable reduction in review overhead.
The timing aligns with rapid progress in both AI coding agents and proof technology. Interactive theorem provers have improved. Automation has advanced. Integration between large models and proof assistants has moved from experiment to prototype. Jane Street intends to experiment with many of these approaches while maintaining its bias toward tools that survive contact with real trading systems.
Not every firm will follow this path immediately. Smaller shops lack the language control or talent density. Regulated entities face additional hurdles around explainability and auditability of proofs. Yet the pressure to verify AI-generated code will only grow. Financial institutions cannot afford subtle bugs in systems that move billions daily. The cost of failure has risen in tandem with system capability.
Minsky closed his post with an invitation. Positions are open in both New York and London. The firm is in early interviewing stages. It seeks collaborators ready to shape the next chapter of reliable software development. For those who have watched formal methods from the sidelines, the signal is clear. A major player in quantitative finance now sees them as essential rather than optional.
The shift does not mean Jane Street has abandoned its pragmatic roots. It continues to invest heavily in testing, type systems and human oversight. Formal methods will complement those strengths, not replace them. The firm’s experience with OxCaml, property-based testing and expect tests provides fertile ground. Agents already benefit from the universal guarantees types supply. Extending those guarantees through proofs could unlock another leap in both productivity and confidence.
Industry observers will watch closely. If Jane Street succeeds in making formal reasoning a daily tool for its engineers, the implications stretch beyond trading floors. Software that powers critical infrastructure, from aerospace to healthcare, could gain similar assurances at lower cost. The combination of human insight, machine generation and machine-checked proof may define the next era of systems that must not fail.
For now the work is just beginning. A new team is forming. Language extensions are under discussion. Integration paths with existing proof ecosystems are being explored. The blog post and job listings mark a public commitment. The real test will come in the months and years ahead, as these ideas meet the unforgiving demands of live markets. Jane Street has bet before on unconventional technology and won. This time the stakes include not only its own systems but potentially the future direction of reliable software at scale.
WebProNews is an iEntry Publication