AI code review tools take one of two approaches: bot reviewers such as CodeRabbit and GitHub Copilot code review leave comments on the pull request, and guided review tools such as Hyperpowers explain the change to the person reviewing it. The third common option uses no AI at all: a person reads the diff on GitHub. Use a bot reviewer when you want a quick automated pass on every pull request, and keep a person as the reviewer when someone has to understand and answer for what gets merged.
They are not exclusive. Many teams run a bot on every pull request and still have a person approve. Here is how each approach works, what it costs and when to use it. We build Hyperpowers, the guided one, so weigh that.
What are the main approaches to AI code review?
| Bot reviewer | Reading the diff on GitHub | Guided human review | |
|---|---|---|---|
| Examples | CodeRabbit, GitHub Copilot code review | GitHub's Files changed tab | Hyperpowers for VS Code |
| Who reads the code | The model | You | You, with a model's explanation beside each change |
| What you get | Comments on lines, and often a summary | The diff, file by file | What the pull request does, then each change in the order the code runs, with what changed and why |
| What gets posted | The bot's comments | Your comments | Your comments, when you press Submit review |
| Where it runs | On the pull request | In the browser | In VS Code and editors built on it |
| Setup | Install the app or turn it on for the repository | None | Install the extension. Creating a walkthrough needs Claude Code |
How do bot reviewers like CodeRabbit and Copilot code review work?
A bot reviewer reads the pull request and posts its findings as comments on the pull request. CodeRabbit installs on your repositories and comments on pull requests, with a summary of the change. GitHub Copilot code review is requested as a reviewer on a pull request and leaves comments, often with suggested changes. GitHub's documentation says Copilot leaves a comment review, and doesn't approve or request changes.
Both review a pull request without taking a person's time. Check each product's current documentation for what it supports, since both change often.
When is a bot reviewer the better fit?
A bot reviewer is the better fit when you have many small pull requests and want each one checked before a person looks. It is also useful when nobody is free to review every change, or when an open source project needs a first pass on outside contributions. It can catch typos, missing null checks and common mistakes early, so a person's review time goes to the parts that need judgement.
What does a bot reviewer leave to you?
It leaves the decision to you. The bot's comments are more text to read and weigh. On a large AI-written pull request, the author may fix the comments with AI too, so the pull request gets changed and the comments get resolved without a person reading the code. Someone still presses Approve. If that person hasn't read the change, nobody has.
What do you get from reading the diff on GitHub?
Reading on GitHub costs nothing and needs no setup. You get every file's diff, comments on lines, suggested changes the author can apply with one click, a Viewed checkbox per file, and a way to see only the changes since your last review.
What you don't get is an order or a reason. Files are listed alphabetically, so a migration, the service that uses it and the page that shows it can be far apart. On a five-file pull request that doesn't matter. On a 70-file one, you spend most of the time working out how the pieces connect.
When is reading on GitHub the better fit?
For small pull requests, it is usually all you need. It also works for anyone with access to the repository, with no editor or extension.
What is guided human review?
Guided human review means a person still reads and approves every change, and a model explains the change first so the reading goes faster. The model writes the explanation. The person writes the comments.
In Hyperpowers: paste a pull request link in VS Code and your own Claude Code explains it. The start page says what the pull request does and how it flows. Then the review takes you through one change at a time, in the order the code runs (database, then backend, then frontend, then tests), with what changed and why next to the real diff in the editor. You comment on lines, and your comments go to GitHub as one review when you press Submit review. Nothing is posted before that.
If the code was written with Claude Code on the machine that created the walkthrough, the why comes from what was asked in those sessions. You can also export a walkthrough as one file, so a teammate reads the same pull request the same way without needing Claude Code.
Hyperpowers can also do a bot-style pass when you ask for one. Review with Claude reads a file, an area or the whole pull request and drafts a comment for each thing worth raising. Each draft waits for you to press Add as comment, Edit or Skip.
When is guided human review the better fit?
It fits when the pull request is too big to read comfortably on GitHub, when it was written mostly by AI, and when your name is on the approval. You end up knowing what the change does, so you can answer for it, fix it, or change it later.
What are its limits?
It takes a person's time, more than letting a bot comment. Creating a walkthrough needs Claude Code and uses your own Claude Code plan. The explanations are written by a model and can be wrong, so each one sits next to the code it describes for you to check. Posting a review needs GitHub. Reading works with any git repository.
Can you use more than one?
Yes. A common setup is a bot reviewer on every pull request for a first pass, and a person who reads the large or risky ones properly before approving. The bot catches routine mistakes. The person makes sure someone understands what is being merged.
Try guided review
Try Hyperpowers in your browser on an example pull request, with nothing to install. To use it on your own pull requests, install the extension and follow Your first review. For the method behind it, see How to review a pull request that AI wrote.
Common questions
Should AI approve pull requests?
That is a team decision, and the tools reflect it. GitHub's documentation says Copilot code review leaves comments and doesn't approve. Hyperpowers never approves or posts on its own. You press Submit review and choose Comment, Approve or Request changes.
Is Hyperpowers a replacement for CodeRabbit or Copilot code review?
It does a different job. Bot reviewers comment on the pull request. Hyperpowers explains the pull request to the person reviewing it. You can use both.
Does Hyperpowers need GitHub?
Reading works with any git repository. Posting your review needs GitHub.
Is Hyperpowers free?
Yes. The extension is free. Creating a walkthrough uses your own Claude Code plan, like any other Claude Code session.