When an AI agent generates a smart contract, reviewing every line of code can become a bottleneck. An article published on November 17, 2026, proposes shifting some of that review effort to the specification: first state in plain English what the change must guarantee, have people review those requirements, and then use formal verification tools to check the code against them.
The article describes a demonstration by Certora CEO Mooly Sagiv involving an ERC-4626 USDC vault. An agent inferred requirements and generated Solidity code and CVL rules, which express properties for Certora’s prover to check. In the demonstration, the prover found a dilution flaw of about 5 wei caused by high-water-mark rounding; the flaw was then fixed.
The point is not that generated specifications can be trusted without review. The article identifies writing the right specification as the hardest part of formal verification and raises a question for teams adopting this workflow: how will they check the quality of the requirements before relying on a passing proof?

