LLM-generated code is fast, but not verifiable. Certora's solution: agents generate formal specifications, humans review them, then use formal verification to guarantee agent-generated code satisfies the specs. Coming to Soroban.
Smart contract development faces a new challenge: AI agents can generate code faster than humans, but verifying agent-generated code is harder. Certora addresses this through formal specification: agents first generate specifications in a declarative language, humans review these specs to ensure they capture intended behavior, then formal verifiers automatically generate code that provably satisfies those specs. This approach guarantees code correctness through mathematical proofs rather than relying on agent correctness alone. Certora is implementing this for Soroban smart contracts, combining AI-driven productivity with formal verification's mathematical certainty. Developers must invest effort in clear specifications, but gain both safer code and automatically generated comprehensive test cases.
