Reproducing Programs With AI Without Knowing How to Build Them: What Is 'REA'?

What happened

An open-source AI tool called REA (Reverse Engineer Anything) has become popular with developers. It helps AI study how a finished program works, even when the source code is not shared. The AI can then help write new code with similar features. REA works with tools like OpenAI's Codex and Anthropic's Claude Code.

Key numbers and facts

• REA passed 300,000 stars on GitHub on the 17th and reached 320,000 on the 19th.
• About 200,000 stars were added in just one week.
• In one example, a rebuilt 1996 Windows game (DX-Ball) gave the same results in 3,205 tests.
• A 1990 game (Ski or Die) snowball fight mini-game was rebuilt by a Finnish developer, but some features like sound were not done.

Background: what is reverse engineering?

Reverse engineering means taking a finished program and working backward to learn how it works. Companies usually keep their source code secret, so users can use a program but cannot see how it works inside. Older reverse engineering tools, such as Ghidra and Hopper, were hard to use without expert skills. REA connects these tools to AI, so a person can describe what they want in normal language.

Why it matters

Free AI-made copies of popular software could compete with paid programs. After HSBC lowered its rating on Adobe and cut its target price, investors showed they see this as a risk. Still, the tools are not perfect. They can make mistakes, they cannot fully copy big programs, and copying protected code can break copyright law. The REA team says results must be checked before anyone trusts them.

What to watch

• Whether companies like Adobe and Microsoft change prices or features in response.
• How copyright rules handle AI-based analysis and copying.
• Whether the rebuilt programs get better at handling complex tasks like image editing and server features.
• How many developers use REA and what new projects appear.