One Developer Used AI to Clone Photoshop, Premiere, and Now Microsoft Office
For years, open-source alternatives to Adobe and Microsoft’s core creative and productivity software existed but never quite matched the originals. GIMP never replaced Photoshop in professional workflows. LibreOffice never closed the gap on Excel’s formula engine or PowerPoint’s layout tools. The engineering cost of matching those codebases was prohibitive.
Brandon Thomas appears to have changed the math.
The software engineer used Claude Opus 5.5 and AI-assisted code generation to build ArtCraft — a suite of open-source Rust implementations of Adobe’s Creative Suite: PhotoCraft (Photoshop), FilmCraft (Premiere), and LightCraft (Lightroom). After the project went viral across GitHub and Reddit, he extended the effort to Microsoft Office, releasing WordCraft, DeckCraft, and GridCraft as free equivalents of Word, PowerPoint, and Excel. The projects were built using techniques the broader developer community now calls vibecoding: directing AI systems to generate production-ready code at a pace no human team could previously sustain.
The Scale of What Changed
The traditional barrier to cloning mature proprietary software was not interface design — it was the decades of accumulated engineering work underneath. Photoshop’s brush engine, Premiere’s timeline model, Excel’s formula calculation tree: each represents years of specialist effort tuned to edge cases that only emerge when millions of users push a product hard.
AI code generation does not eliminate that complexity. What it does is compress the time required to produce a first working approximation. A developer who previously needed a team of twenty and three years to ship a credible alternative can now produce one alone in months. Thomas is the evidence.
What This Means for the Software Industry
The economic argument for proprietary productivity software has always rested on two pillars: the cost of development and the network effect of compatibility. AI vibecoding attacks the first pillar directly. If one engineer can replicate five years of Adobe engineering in a few months, the marginal cost of a viable competitor drops to near zero.
The compatibility pillar still holds. A vibecoded Word clone that cannot reliably import a complex .docx with tracked changes and embedded objects is not a real substitute for most enterprise users. The question is how long that gap persists before AI-generated file format handling closes it too.
Open-source software has historically succeeded when it achieved rough functional parity with proprietary alternatives and removed the cost barrier entirely. What is new here is how fast parity is now reachable. The gap between “nobody could afford to build this” and “one person built this over a weekend” is compressing.
What Comes Next
The immediate signal to watch is adoption. Vibecoded clones that gain active contributor communities can improve quickly; ones that do not tend to stall at the prototype stage. Thomas’s Adobe clones attracted enough attention to motivate a Microsoft Office follow-on. Whether either project generates the pull-request volume needed to harden edge cases determines how much competitive pressure it actually applies.
For enterprise software vendors, the threat is not that a vibecoded clone displaces them in large accounts next quarter. It is that the willingness to pay for software that previously had no credible free alternative now has a ceiling, and that ceiling just got lower.