Tag1 Guide Explains Human-Controlled AI Code Reviews

AI-generated code review comments can be kept unpublished until a developer checks them, according to a guide by Jeremy Andrews, founding partner and CEO of Tag1 Consulting. Published on 30 September 2026, the guide explains how an AI assistant can stage comments in GitHub pending reviews or GitLab draft notes for a person to edit, discard or submit under their own account.

The approach differs from automated review bots that publish findings directly or use results to gate merges. Andrews describes a four-step sequence: the agent drafts comments, the reviewer checks them in the usual interface, the agent may optionally review those edits, and the person submits the final review. The guide also links to a prebuilt review tool by Greg Chaix that can stage findings for human approval.

The GitLab section is particularly relevant to Drupal contributors reviewing merge requests on DrupalCode. It shows how to authenticate the glab command-line tool against git.drupalcode.org and create draft notes through the GitLab API, including the diff references needed for inline comments. GitHub users can similarly work with pending reviews through the GitHub API or Model Context Protocol tooling.

Andrews recommends telling the agent not to submit, approve or publish a review and using the human reviewer's credentials so the draft appears under that person's account. Those instructions are workflow safeguards, not independently enforced limits on an agent's permissions; access to credentials and review-submission functions still requires care. The guide provides commands and a practical pattern, but does not present comparative evidence that AI-assisted reviews are faster or more accurate.