AI coding agents can help developers turn observations about an application into a new implementation—and help interpret evidence from software-analysis tools. ArtCraft’s Adobe-style apps are a prominent example of the first approach. They demonstrate how quickly a small team can generate a substantial codebase, but not that the result is a reliable replacement or a copy of the original source code.
4
5
What ArtCraft built—and how
Developer Brandon Thomas used Anthropic’s Claude Opus 5.5 to help build free, open-source Rust applications modeled on Adobe Creative Cloud software. The early suite included PhotoCraft, LightCraft, FilmCraft and EffectCraft, aimed respectively at Photoshop, Lightroom, Premiere and After Effects.
5
13
34
Thomas cited frustration with Adobe’s cancellation fees as one motivation for the project. Reporting says those fees can reach half the remaining annual balance.
5
The project’s stated approach is a “clean-room” reimplementation: build new software based on public specifications and observed behavior, rather than copying the original code. That is a description of the development method, not proof that the apps match Adobe’s features or performance.
5
The scale of AI-assisted development can be striking. XDA Developers reported that EffectCraft reached roughly 330,000 lines of Rust about 10 days after its first commit, with four to eight agents reportedly working at once.
38 Another report describes the team’s estimate of EffectCraft’s professional readiness as about 45%.
32 These are reported project figures and estimates, not independent measures of software quality.
ArtCraft and REA are different projects
Reverse Engineer Anything (REA) connects coding agents to tools such as Ghidra and IDA, which can help analyze software. That makes it a bridge between an AI agent and analysis tools—not, by itself, an automatic application-cloning system.
4
11
ArtCraft’s stated clean-room method is distinct: its Adobe-style apps are not described as products of REA or as decompilations of Adobe software.
1
4 Keeping that distinction clear matters. Rebuilding an application based on its visible behavior and analyzing compiled internals are related ways to learn about software, but they are not the same process.
ArtCraft has also presented Craft Anything, a separate open-source project. Its repository describes a workflow that studies a website, app or game, has a coding agent build a new project, then compares and refines the result.
18 The project’s description sets out an intended workflow; it does not establish that the engine can reliably recreate arbitrary software.
Fast code generation is not the same as readiness
ArtCraft’s speed should be weighed against its maturity. PhotoCraft has been described as early-alpha software and “not yet a Photoshop replacement.”
4 The reported 45% professional-readiness figure for EffectCraft is also a project estimate, not a guarantee of performance in real production work.
32
The Office-style apps that followed—including WordCraft—are likewise not established replacements for Microsoft 365.
6
13 Large code counts can signal rapid implementation, but they do not, on their own, establish usability, compatibility, stability or completeness.
What the example does—and does not—settle
ArtCraft points to a lower barrier to building software that resembles familiar commercial tools. But its early-stage status leaves the practical test unresolved: whether the apps can handle real users’ workflows reliably over time.
The project’s clean-room claim is relevant to how its developers say they worked, but it is not a legal ruling. The available reporting does not establish a court decision about ArtCraft’s approach.
5
There is also a community question beyond commercial software. Decomp.dev, a site that tracked game-decompilation projects, disabled its main public index and stopped accepting new projects while keeping existing project pages and historical data available. Its administrator said AI had “broken the community aspect” of decompilation.
47
59
The broader lesson is not that software has become effortless to replace. AI agents can accelerate implementation and change who can attempt it; reliability, legal questions and the health of the communities doing this work remain separate challenges.