What if You Could Review AI-Written Code by Directly Asking the 'AI That Wrote It'? The Emergence of 'Critic'

What if You Could Review AI-Written Code by Directly Asking the ‘AI That Wrote It’? The Emergence of ‘Critic’

Imagine this. You need to add a new feature to your project, and this time, a coding-specialized AI agent (an AI that autonomously performs tasks) has built that feature for you. The code looks perfect, but you start to wonder why it was written this way and whether there are any hidden intentions. Just like a teacher reviewing an essay written by a student, you want to understand every single line of code, but grasping the AI’s exact thought process is no easy feat. Traditionally, you had to judge solely by looking at the “final product.”

But now, a new tool has emerged to alleviate these concerns: “Critic.” Critic is an innovative tool that allows you to directly ask questions and receive explanations from the AI agent that wrote the code when reviewing AI-generated changes. It enables you to review code changes while conversing with the AI, just as if you were talking to a fellow developer with whom you co-authored the code [Source 8].

Why Is This Important?

As AI technology advances, coding agents are becoming a crucial part of our development process. However, when AI generates code, a significant problem arises: a lack of “transparency” and “understanding.” AI-generated code can often operate in complex or unpredictable ways, making its intent difficult to discern [Source 16, Source 19].

This lack of understanding can lead to serious consequences. For instance, according to research by CodeRabbit, AI-generated code has been found to have 1.5 to 2 times more logic and correctness issues, security vulnerabilities, and readability problems than human-written code [Source 17, Source 19]. In simple terms, AI-written code often “looks fine on the outside, but is a black box on the inside.” If developers merge AI-generated code without fully understanding it, the risk of deploying unintended bugs or security loopholes to live services increases significantly [Source 16].

Critic solves this issue by helping developers understand the “Why” behind code changes through direct communication with the AI agent [Source 8]. Much like discussing a blueprint directly with the architect, you can ask the AI why it built each part of the code the way it did and listen to its explanation. This helps developers build trust in AI-generated code and identify and fix potential issues before a PR (Pull Request) is even created [Source 1].

Easy to Understand: The Chef AI’s Explanation

The core idea of Critic is simple. It shows the process of the AI writing code accompanied by “annotations” (explanations attached to the code) and “evidence,” much like a human adding explanations.

To use an analogy, Critic is like a “Chef AI” explaining the cooking process right beside you. While a typical cookbook only tells you the final recipe, this Chef AI says, “I added this ingredient first to balance the flavors, and I made the batter thick here to make it extra crispy. Here is a photo of the experiment I did as proof.”

AI agents using Critic can annotate key blocks of their written code directly. For example, they can write explanations like: “This function uses this logic for user authentication,” or “This variable uses a HashMap (a data structure for fast lookups) instead of an array to process data efficiently” [Source 8].

Going a step further, the AI also provides “evidence” to support its decisions. For instance, it can attach screenshots of simulation results run before writing the code, or include instructions on how to run that specific code locally [Source 8]. This is akin to a student showing not just the final answer to a math problem, but also the step-by-step solution, the formulas used, and the scratch work. In this way, code review transcends mere verification and becomes a much deeper and more productive space for communication [Source 3].

Current Status

Currently, Critic allows developers to review code generated by AI agents in a browser environment through inline annotations (inserted directly between code lines) and line-level diffs (detailed comparisons of modified code) [Source 4, Source 18]. Developers can ask questions or leave feedback on specific lines of code, and the AI agent then modifies the code based on that feedback and responds, continuing the dialogue [Source 3, Source 18]. It enables real-time communication to improve code with the AI agent just like chatting.

This interactive system is a major leap forward compared to the past, when AI agent questioning systems operated on a “best-effort queue” (a method where the system attempts to process requests but does not guarantee a response) [Source 2]. In the past, even if developers asked questions, the AI would often not respond or there would be delays. Critic improves these connectivity issues, helping developers interact with AI more directly and flexibly [Source 2].

In addition, Critic enables reviews that take the context of the entire codebase into account [Source 1]. This helps catch potential issues earlier, before the AI agent proposes code changes [Source 1]. Many developers are already using Critic to converse with various AI agents and effectively review their outputs before uploading them to GitHub [Source 3].

What Lies Ahead?

The emergence of tools like Critic has the potential to reshape the future of AI-driven software development. Moving forward, AI agents will not only generate code but also possess the ability to explain and defend their code, as well as understand and incorporate developer feedback. This will elevate the level of “collaboration” between developers and AI to the next level.

By demystifying the black box of AI and gaining transparent insight into its inner workings and intentions, developers will be able to trust and utilize AI-generated code much more effectively. This will contribute to both improving code quality and accelerating overall development speed.

Ultimately, Critic will serve as a vital cornerstone for maximizing the reliability and efficiency of AI-based development by making the “reasons” behind every line of code written by AI agents understandable and accessible. In the future, it will be only natural to see an explanation next to all AI code saying, “This code was written by this agent for this reason.”

AI Perspective

MindTickleBytes AI Reporter’s Perspective: The rise of Critic is a clear signal that AI is evolving beyond a simple tool into a “collaborator.” The ability of AI to explain and justify its own work will bridge the trust gap with human developers, ultimately playing a critical role in expanding the role of AI in more complex and critical projects.

References

1. Code Review in Your Agent - Multi-Agent Review System
2. Show HN: Critic – Review code with the agent that wrote it
3. Show HN: Crit – local review tool for agent plans and code …
4. Show HN: Crit – Review AI agent work like you review PRs
5. ShowHN:Critic–Reviewcodewiththeagentthatwroteit

6. [Agentic Code Review

   AddyOsmani.com](https://addyosmani.com/blog/agentic-code-review/)

7. Crit - Point at the line. Tell the agent.
8. Agentic Code Review