---
title: "Our Framework for Reviewing AI-Generated Code"
description: "JetBrains Research, in collaboration with Lund University, studied how to review AI-generated code as AI agents produce thousands of lines across multiple files quickly. The core problem identified is trust calibration: developers cannot gauge an LLM’s confidence per line, making line-by-line review unscalable. The study proposes a three-level conceptual framework: overview for high-level hypotheses, file-level risk stratification, and code snippet-level chunk decomposition with chain-of-thought linkage. This framework was developed through participatory design with 17 practitioners and a follow-up survey of 43 professionals, and will be presented at ESEIW 2026. The paper argues that current diff viewers are inadequate for AI-generated code and that tools must surface risk and confidence signals at the granularity where developers allocate attention."
---

# Our Framework for Reviewing AI-Generated Code

[Jetbrains](https://yomu.fyi/company/jetbrains) · Katie Fraser · Oct 6, 2026

**Type:** Explainer

## Summary

JetBrains Research, in collaboration with Lund University, studied how to review AI-generated code as AI agents produce thousands of lines across multiple files quickly. The core problem identified is trust calibration: developers cannot gauge an LLM’s confidence per line, making line-by-line review unscalable. The study proposes a three-level conceptual framework: overview for high-level hypotheses, file-level risk stratification, and code snippet-level chunk decomposition with chain-of-thought linkage. This framework was developed through participatory design with 17 practitioners and a follow-up survey of 43 professionals, and will be presented at ESEIW 2026. The paper argues that current diff viewers are inadequate for AI-generated code and that tools must surface risk and confidence signals at the granularity where developers allocate attention.

## Context

AI coding agents can now generate large amounts of code quickly, but reviewing that code is difficult because LLMs present all lines with equal confidence, lacking the social and contextual cues of human-authored code. This creates a trust calibration problem where developers cannot efficiently allocate review effort.

## Approach / What changed

A three-level conceptual framework for reviewing AI-generated code: overview for high-level hypotheses, file-level risk stratification, and code snippet-level chunk decomposition with chain-of-thought linkage. Developed through participatory design with 17 practitioners and a survey of 43 professionals.

## Takeaways

- The core problem with reviewing AI-generated code is trust calibration: developers cannot gauge an LLM’s confidence per line, making line-by-line review unscalable.
- The proposed framework follows Shneiderman’s information visualization mantra: overview first, then zoom and filter, then details on demand, mapped to three levels of review.
- Current tools like CodeRabbit, Claude Code, and Graphite offer partial counterparts but none fully address trust calibration or impose the three-level workflow.

## FAQ

### Why is reviewing AI-generated code different from reviewing human-authored code?

Human-authored code comes with social and contextual cues like the author's seniority and confidence, while LLMs present every line with the same apparent confidence, removing the invisible scaffolding that helps reviewers allocate effort.

### What is trust calibration?

Trust calibration is the capacity to allocate review effort proportionate to segment-level risk when the author cannot be interrogated about their confidence or reasoning. It is the skill of knowing where to look hard and where to skim.

### What are the three levels of the proposed review workflow?

The three levels are overview (high-level hypotheses), file-level risk stratification, and code snippet-level chunk decomposition with chain-of-thought linkage, following Shneiderman's overview-first mantra.

**Tags:** [AI](https://yomu.fyi/topic/ai), [AI Agents](https://yomu.fyi/topic/ai-agents), [Code Review](https://yomu.fyi/topic/code-review), [Trust Calibration](https://yomu.fyi/topic/trust-calibration)

- Source: [Jetbrains](https://blog.jetbrains.com/research/2026/10/review-ai-generated)
- Source URL: https://blog.jetbrains.com/research/2026/10/review-ai-generated
- Ingested by Yomu: 2026-10-06T15:00:10.345Z

[Read original post](https://blog.jetbrains.com/research/2026/10/review-ai-generated)