Reviewing AI-Generated Solidity by Checking the Specification First

When an AI agent generates Solidity, reviewing every line of code can become a bottleneck. An article published on November 17, 2026, proposes starting with a different review task: write down the properties a change must satisfy in plain English, have a person 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. In the demonstration, an agent inferred properties for the vault and generated both Solidity code and CVL rules for verification. The prover found a dilution issue of about 5 wei caused by rounding of the high-water mark; the issue was then fixed.

That example illustrates both the potential and the limit of the approach. A prover can check code against stated rules, but the team still has to decide whether those rules capture the intended behavior. The article identifies writing the specification as the hardest part of formal verification and discusses checking specification quality and beginning a team practice of spec review.