This Is What Happens When AI Reverse-Engineers Corporate Software

Summary

The video examines claims that AI agents have helped create open-source, “clean-room” reimplementations of Adobe software, including PhotoCraft (Photoshop-like), FilmCraft (Premiere-like), and LightCraft (Lightroom-like). The projects are reportedly written primarily in Rust and available on GitHub.

The presenter describes them as examples of AI-assisted black-box reverse engineering: agents study public documentation, file formats, screenshots, tutorials, and software behavior, then build separate implementations without using the original source code. Work can be divided among agents—for example, by feature—and checked through repeated automated testing.

For PhotoCraft, the video highlights a shared registry of more than 500 commands used by the UI, command line, JSON control channel, and MCP server. It also describes tests that compare PhotoCraft’s output with results produced by Photoshop.

The projects are still incomplete, estimated at roughly 60–70% of the original functionality, with broken workflows, missing tools, and substantial testing still needed. The video presents them as an early indication of what may be possible, not as fully equivalent replacements.

Key Analysis

• Potential impact on proprietary software: If AI makes it cheaper to reproduce established features and workflows, software development may become less of a competitive moat. Businesses could face more competition, pricing pressure, and less vendor lock-in.
• Why open source could benefit: Compatible alternatives could make it easier for users to move away from proprietary toolsets and may lower barriers to building software.
• Limits to the threat: Adobe may retain advantages in integrations, services, brand trust, and professional support. Unfinished alternatives may also be risky for users.
• Legal uncertainty: A clean-room approach aims to reproduce behavior without copying protected source code, but the video raises unresolved questions about AI training data, similar-looking interfaces, confidential information, and the distinctions among copyright, patents, trademarks, and trade secrets.

Reviews, Guides, and Tutorials

• The video is an analysis and repository walkthrough, not a step-by-step tutorial or product review.
• It points viewers to the GitHub projects and discusses their claimed capabilities, architecture, and current limitations.

Main speaker/source: SavvyNik, the video’s presenter. The projects discussed are attributed in the subtitles to the Arc Craft/ArtCraft project and its GitHub repositories.