What do you actually look for in the first 60 seconds of a PR review? (Specifically for AI-generated PRs)

I’m currently working on a pipeline to audit code generated by autonomous AI agents (essentially an "anti-hallucination" trust gate before merging). Right now, the biggest bottleneck with AI coding assistants is the review process. They generate massive walls of text, dump repetitive bot logs, and leave reviewers with a huge cognitive load. You often spend more time figuring out *what* the AI actually did than reviewing the code itself. I want to build a system that intercepts these PRs and generates a highly readable, high-signal "Review Artifact" that gives human reviewers exactly what they need right at the top. To make this actually useful, I’d love to hear how you handle your raw PR workflow: 1. **The First 60 Seconds:** When you open a PR, what exactly are you scanning first to gauge the blast radius and risk? 2. **Signal vs. Noise:** How do you quickly separate the critical stuff (auth, DB schema changes, dependency bumps) from the noise? 3. **The "Trust" Evidence:** If an AI agent wrote the PR, what specific *evidence*, guarantees, or summary would you demand to see in the description to actually trust its output and speed up your review? Feel free to roast the worst AI-generated PRs you’ve had to deal with. I want to know exactly what formatting or info actually reduces your mental load. Thanks!