The app can be copied without its source code #
Video creator Matthew Berman recently posited a provocative idea: AI could allow developers to rebuild complex software like Photoshop by observing its behavior, rather than accessing its proprietary source code. This approach bypasses traditional reverse engineering, which often involves decompiling binaries.
The core loop involves AI agents inspecting an application's screens and workflows, then generating an implementation. These agents, often multimodal LLMs like Claude 3.7 Sonnet, compare their output with the original, refining the result through iterative vision-to-code diffing until it achieves functional and visual parity. Pixel-perfect visual similarity is merely one metric; functional equivalence drives the process.
This method, dubbed "clean-room" AI reverse engineering, avoids direct code copying. Projects like morluto/rea, a trending open-source TypeScript CLI and MCP server on GitHub, automate this process, connecting coding agents to local analysis tools. This framework allows developers to map and recreate functionality without illicitly obtaining source code.
While this technique sidesteps direct source code theft, the legal implications remain nuanced. Clean-room methods and interoperability have legal precedents, but outcomes depend heavily on the specific methods employed, the extent of copied functionality, and the jurisdiction. The line between legal reimplementation and intellectual property infringement remains a dynamic frontier.
REA gives coding agents a toolbox #
Rebuilding complex software without source code sounds like magic, but projects like reverse engineer (REA) illuminate the practical steps. REA, an open-source framework highlighted in the video, lives on GitHub at github.com/morluto/rea, complete with documentation outlining its approach. This isn't a one-prompt solution; it’s a sophisticated toolkit for coding agents.
REA connects AI agents to a suite of software-inspection tools and workflows. These agents don't instantly clone an application; they methodically observe, analyze, and rebuild. This process involves integrating local analysis tools like Ghidra, Hopper, and IDA Pro, alongside Electron package extractors and dynamic runtimes, to map functionality.
The idea of visual-to-code generation isn't new; NVIDIA has demonstrated turning UI designs into functional code. However, recreating a static screen differs vastly from reconstructing a complex application’s internal behavior and intricate state transitions. REA tackles this by employing multimodal LLMs, like those with advanced vision capabilities, to perform iterative screenshot-to-code diffing.
An agent takes a UI render, compares it against the target application, generates code, and repeats the cycle until it achieves a "pixel-perfect" match. This iterative, observation-driven approach systematizes the reverse engineer’s craft, moving beyond simple UI elements to tackle deeper functional replication.
A Photoshop look-alike isn’t Photoshop #
Interface fidelity is one thing, but production capability is another entirely. Photoshop’s decades of refinement in color science, GPU rendering pipelines, and robust file compatibility are far more complex to replicate than menus and panels. AI can reproduce a visual shell, but the underlying precision needed for professional creative work demands a depth of engineering that goes beyond pixel-perfect imitation. Yet, an AI-made subset of Photoshop, focused on a user’s specific needs, offers compelling possibilities. Imagine a lightweight editor built around your three essential tasks, stripped of bloat and telemetry, and enhanced with bespoke features a general-purpose suite lacks. This is the promise of hyper-personalized software: a tool designed precisely for you, not the average user.
Matthew Berman’s video highlights game mashups like Mario in Elden Ring or Minecraft in GTA to illustrate this appeal—combining familiar elements in novel ways. While creative software also demands dependable, precise results, the concept of unbundling and recomposing tools offers a powerful vision. This approach allows developers to reverse engineer specific functionalities, rather than cloning entire applications, creating custom experiences. For further exploration of this agentic reverse engineering, check out the open-source project GitHub - morluto/rea: reverse engineer with agents.
Enjoying this? Get one like it in your inbox each morning.
one email a day · unsubscribe in two clicks · no third-party tracking
The real threat is software made just for you #
Claiming people will "never pay for software again" stretches credulity. Custom applications, even AI-generated ones, still demand significant compute resources, developer time, rigorous testing, and continuous updates and maintenance. The notion that an AI-built app is "basically free" overlooks the entire lifecycle cost of software.
Legally, the landscape remains complex. While reverse engineering for interoperability is often permissible, replicating a distinctive user interface, reusing proprietary code, or copying branding elements can trigger different intellectual property challenges. An AI-generated implementation offers no blanket legal shield against claims of copyright or trademark infringement.
The more credible shift lies in the democratization of small, purpose-built tools. AI agents could make creating hyper-specific utilities cheap enough to generate on demand, tailored to individual workflows. Imagine an agent building a tool that extracts specific data from images, then automatically uploads it to a custom database—a task too niche for a commercial suite.
This could pressure the existing subscription bundle model, forcing vendors to justify the breadth of their offerings. While comprehensive professional suites like Adobe Photoshop or Adobe Adobe Premiere Pro Pro will likely remain difficult to fully replace due to their deep feature sets and mature ecosystems, the market for bespoke, on-demand tools could explode. We are shifting from one-size-fits-all to software made just for you.
Frequently Asked Questions #
What is AI-driven software reverse engineering?
It uses AI agents to inspect an app’s interface and behavior, then build a separate implementation that aims to reproduce selected features.
Can AI really recreate Photoshop?
AI can reproduce visible interfaces and simpler workflows, but matching Photoshop’s mature rendering, color science, performance, and reliability is far harder.
Is it legal to make an AI-generated clone of an app?
It depends on how the work is done and what is copied. Reimplementing functionality may differ legally from copying code, assets, branding, or protected visual design.
What is the Reverse Engineer Anything project?
REA is an open-source project that connects coding agents with tools for inspecting and analyzing software. Its capabilities and legal implications depend on how it is used.