Craig Hughes

Reasonable

As AI agents generate code faster than humans can review it, the bottleneck has shifted from writing code to trusting it.

Formal verification solves this, but resisted automation for decades: fundamentally hard, demanding mathematical rigor that even current LLMs cannot produce reliably.

Reasonable trains models to reason formally about software, use complex formal programming languages, and provide verifiable correctness guarantees for software - abstracting away the complexity that previously slowed adoption.

Pushing software development deep into superhuman territory requires an independent layer of verification and trust.

Reasonable is building that platform layer.