As models become smarter and coding tools become better at harnessing that power for larger and longer tasks, two things are becoming clear:

1. It’s more important than ever for software engineers to understand what their code does and how it works. “Vibe coding” is great for tiny codebases and prototyping, but without understanding what’s going on under the hood, you will drown in technical debt, compromise your ability to direct your AI minions effectively, and lose the respect of your peers–if you’re lucky enough to avoid catastrophic failures in production.
2. Today’s AI coding tools push you hard in the direction of YOLOing it. This is a product of affordances and impedance mismatch: it’s easy to write code (or to cause it to be written) with a glorified chat box, but much harder to effectively review and understand it.

A 10x Solution for Code Review

Effective AI-assisted code review requires three things: Context, to keep the AI from flagging problems that don’t actually exist; a UI that allows you to easily focus on the most relevant changes; and an understanding of the intent behind the changes.

Brokk’s philosophy is to keep you, the human supervisor, in the loop as much as possible, and the way we tackled these problems in Guided Review is no exception. Brokk isn’t looking to replace you as a reviewer so much as help you use your time effectively, on the most important changes, while updating your mental map of what’s going on in your code.

Thus, Guided Review deliberately concentrates on hot spots—code paths and design decisions with a high impact on behavior, maintainability, or technical debt—rather than surfacing a long tail of low-value comments on stylistic or mechanical issues.

Here’s what this looks like:

Brokk will give you an Overview, Key Changes, Design Notes, Tactical Notes, and additional Tests. The last three sections each contain Recommendations that you can enqueue as Lutz Mode tasks without leaving the review UI.

Brokk Knows What you Want

Everything you do (and everything your AI does) in Brokk is part of a Session that Brokk saves locally and to the cloud; the Session contains a log of your activity and changes, with full branching and undo/redo support.

When you ask for a Guided Review, Brokk pulls relevant context out of all the Sessions associated with those commits:

1. Your instructions,
2. The Lutz Mode tasks those were turned into, and
3. Any additional context Brokk used while making changes.

Like everything Brokk does, the inputs used for the review are visible in your Context, where they can be manipulated or just used for followup questions.

Including the Session metadata in the review helps Brokk understand what your intent was, and catch problems like incomplete or missing functionality.

Review Anything

You can invoke a Guided Review on your own code, or on a colleague’s.  Technically, this can be any commit range, but most often you will be reviewing HEAD of some local branch or of a pull request against its merge-base. Here’s what reviewing a Pull Request looks like:

2026 will be the Year of Better Reviews

In 2025, LLMs and agents solved the problem of writing code. Now reviewing that firehose of output is the pain point. Brokk can help. Try Brokk for free today!