Your Vibe-Coded App Can Look Right And Fall Apart After Launch. REA Shows You How Real Teams Build.

Summary

The video argues that REA (Reverse Engineer Anything) is not the end of software development. Its value is helping people understand how existing software works so they can turn a general idea into specific requirements and make better implementation decisions. An AI coding agent can write code, but the builder still needs to know what the software should preserve and how it should behave.

REA connects to coding agents such as Claude Code or Codex to inspect software on a computer and analyze available code. The presenter describes it as an “X-ray” for software: useful both as an accelerator for developers and as a learning tool for people without programming experience. The goal is to learn from existing implementations—not simply clone them—and adapt relevant behaviors to a new product.

Examples and analysis

• Aegis authenticator: An analysis traced how an authenticator generates its six-digit codes by combining an account key with the current time in 30-second intervals. Changing either input helps explain why the code changes.
• Notion copy and paste: The investigation showed how copied content can include ordinary text and HTML alongside structured block data and identifiers that let Notion reconnect pasted content to its richer structure. This illustrates why requirements such as preserving a source link, notes, or other relationships need to be specified—not just “make copy and paste work.”
• Windows Calculator: The percent button behaves differently depending on context: 200 + 10% gives 220, while 200 × 10% gives 20. The example shows how inspecting a working implementation can reveal behavior that a simple feature description or screenshot would miss.
• Ledger Reborn and QuickBooks Desktop: A community developer used REA to investigate accounting behavior—including inventory calculations, reconciliation, and reporting—and then built an independent Python and SQLite application. The presenter emphasizes that understanding calculations and workflows matters more than copying an interface.
• Chromium dinosaur game: An investigation traced the game’s increasing speed to its code: velocity starts at six and increases slightly on each update, up to a limit. That rule could then be adapted in a separate game.
• DX Ball: A case study traced how a brick’s horizontal position affects the sound’s left-right panning. The example shows that a focused investigation can uncover a specific behavior without reconstructing an entire program.

Practical guidance

The presenter recommends beginning with the work the software should support, then using the agent to investigate comparable components and ask targeted questions. For example, ask what information should be retained when copying, what happens when associated data is deleted, or how a feature handles different cases. The answers can affect data models, editor choices, storage, validation, and the technology stack.

Keep investigations narrow: tell the agent what behavior to trace and how much detail is useful, rather than asking for an exhaustive explanation of a whole application. Developers can also specify technical constraints and ask about dependencies or implementation options. The presenter notes that setup may involve tools such as Node.js and npm, and suggests asking an AI agent whether optional additions are needed for a particular project.

REA can only inspect what is available to it. The video cautions that closed server-side code may not be visible; analyzing app behavior may offer clues, but attempting to access private backend code is not appropriate. The presenter’s main recommendation is to use REA for learning and to verify that the resulting design supports the actual workflow.

Main speakers and sources

• Nate B. Jones, the presenter and commentator.
• REA case studies and investigations discussed in the video: Aegis, Notion, Windows Calculator, Chromium’s dinosaur game, and DX Ball.
• A community developer’s Ledger Reborn project, based on investigation of QuickBooks Desktop.