A review of 50 documents from 31 organizations finds that the standard code review process, where developers propose changes and a human approves them before the code goes live, was designed for human authors. Agents produce code faster and in higher volume than that process can handle. Reviewers end up approving work they cannot fully judge. The record of who is responsible for a change becomes unclear. The review surfaces two concrete problems. First, accountability gaps: when an agent writes code and a human approves it, neither party owns the decision the way the old workflow assumed. Second, dataset blind spots: the training data used to build coding agents comes from existing repositories. Agents learn patterns from human-paced, human-authored work that does not reflect how agents produce code. The paper does not name a replacement workflow. For designers building tools where agents and humans share a codebase, the practical question is what the human needs to see, and decide, before approving work the agent produced. The review suggests the current answer, a summary of changes and a comment thread, is not enough.