# New LLM-assisted methodology improves hardware and software reverse engineering

> **Key takeaway:** A new workflow uses AI coding agents equipped with specialized skills and debugging loops to automate the analysis of binary code and hardware systems.

**Published:** 2026-11-19T22:17:09+00:00 
**Source:** Adafruit Electronics 
**Category:** open-hardware 
**Canonical URL:** https://fosswire.org/news/new-llm-assisted-methodology-improves-hardware-and-software-.html 

## Summary
A researcher known as ioma8 has detailed a new methodology for utilizing Large Language Model (LLM) coding agents to accelerate the reverse engineering of hardware and software. By providing agents with specific, purpose-built skills, the process can generate high-level code overviews from raw assembly much faster than manual analysis.

The approach addresses common failures in AI-driven analysis, such as agents getting stuck in repetitive loops or failing to execute disassembly steps. The methodology introduces structured tools for interacting with existing reverse engineering software and implements a rigorous debugging protocol to ensure accuracy.

## Technical components of the AI-driven workflow
- **Tooling Integration:** The workflow utilizes specialized skills to bridge AI agents with established reverse engineering tools like radare2 and the MCP bridge for Ghidra, including 22 custom Ghidra scripts for targeted decompilation.
- **Hypothesis-Driven Debugging:** To prevent random guessing, the system employs a 'hypothesis-debugging' skill that forces the agent to state a single testable hypothesis and log results in a CSV loop, mimicking the scientific method.
- **Automated Documentation:** The agent is configured to output all findings and analytical progress directly into Markdown files during the investigation process.
- **Disassembly Automation:** Specific skills are provided to ensure the agent can successfully navigate the disassembly phase, a step where standard agents frequently fail.

## Why it matters
This methodology provides a blueprint for security researchers and hardware engineers to use AI agents more effectively for auditing closed-source firmware and binaries. By integrating with open-source tools like radare2 and Ghidra, it strengthens the capability of the open-source community to analyze and understand proprietary hardware and software systems.