AI Code Liability Falls on Whoever Approves the Merge - Automated Verification Is Non-Negotiable
When AI-generated code causes an outage, the conversation immediately turns to whether AI is ready for production. That is the wrong question. The right question is whether your verification process is ready for the volume of AI-generated code you are shipping.
Liability Lands on the Approver
The legal and professional reality is clear - whoever approves the merge owns the outcome. It does not matter if a human or an AI wrote the code. The person who clicked "approve" on the pull request is responsible.
This is not new. We have always been responsible for code we approve, even when a junior developer wrote it. What is new is the volume. When AI agents can produce 10x more code per day, the review bottleneck becomes the liability bottleneck.
Why Manual Review Does Not Scale
You cannot manually review every line of AI-generated code at the pace agents produce it. The math does not work. If an agent generates 500 lines of code in 15 minutes and a thorough review takes 30 minutes, you have eliminated the productivity gain.
The answer is not to skip review - it is to automate verification. Build systems that test the code the same way a human would evaluate it.
Automated Verification in Practice
The approach that works - the agent writes code, then tests it by running the app, navigating to the relevant screen, and verifying the output. For UI changes, it takes screenshots and compares them. For logic changes, it fires test hooks and checks results.
This is not unit testing (though that helps too). This is end-to-end verification where the agent actually uses the feature it just built and confirms it works correctly.
The workflow looks like: write code, build the app, launch it, navigate to the feature, verify the behavior, check logs for errors. If anything fails, fix it and try again. Only mark the task as complete after verification passes.
The Investment That Matters
Building automated verification into your AI coding workflow is not optional - it is the single most important investment you can make. Every minute spent on verification infrastructure saves hours of incident response and protects you from the liability that comes with shipping unverified code.
Fazm is an open source macOS AI agent. Open source on GitHub.