Ivan Gavran Reopens Formal Verification Debate

Formal verification is not a magic correctness button, but AI-generated code makes its feedback loops increasingly hard to ignore.

• –The hardest problem remains specifying what “correct” means; proving the wrong requirement only creates false confidence.
• –AI agents make verification more attractive by generating code quickly while leaving humans responsible for intent and oversight.
• –Lightweight specifications, model checking, testing, and targeted proofs are more realistic than verifying entire production systems.
• –The essay’s strongest conclusion is that reliability comes from layered engineering practices, not verification alone.
• –This is especially relevant for critical infrastructure, finance, and agent-generated software where review capacity cannot keep pace with output.