The Story of ArtCraft and What it Means for Software Moats

I've been watching one GitHub org for a couple of weeks now and I can't stop thinking about it.

It's called ArtCraft. And what it's doing tells you a lot about where software is going.

The story of ArtCraft

ArtCraft started as a pretty normal idea. An open-source desktop app for making AI images and video, built for artists who want real control. Not just typing a prompt and hoping. You compose the shot yourself. Pose the characters, place props in 3D, build the scene, then let the model render it.

The pitch was about ownership. Stop renting your tools from websites. No subscription, no middleman. The whole studio is on GitHub. Fork it, keep it, it's yours.

Fine. Lots of open-source tools say that.

Then the org kept going.

Next came PhotoCraft, a clean-room reimplementation of Photoshop in Rust. Then LightCraft, for Lightroom. FilmCraft, for Premiere. PdfCraft, for Acrobat. Then they left Adobe and went after everything else. WordCraft for Word. GridCraft for Excel. DeckCraft for PowerPoint. CADCraft for AutoCAD. SoundCraft for Pro Tools. All open source. All free.

Here's the detail that got me. PhotoCraft, the Photoshop clone, has around 46,000 stars. The original ArtCraft app, the thing the company was actually founded on, has about 16,700. The copy is outpulling the product almost three to one. People don't just tolerate clones. They're excited about them.

The other half: REA

ArtCraft is one org rebuilding the big apps. REA is the tool that lets anyone do it.

REA stands for Reverse Engineer Anything. You plug it into your coding agent. Then you say something like: figure out how search works in this app, show me the evidence, and build a version for my project. It inspects the app without the source code. Native binaries, Electron apps, websites. It explains how the feature works and hands that to your agent to rebuild.

No moat

Put those two next to each other and the conclusion is hard to dodge. With enough money and people, any software can be reverse engineered. Whether the creator likes it or not. The code is not a secret anymore. The feature is not a secret anymore. The moment something is valuable, it's copyable.

That breaks the thing most of us quietly believed was the moat. Code, and the features you figured out by working with customers, that was the moat. Now it's a free sample.

So where does the value go

If a feature can be copied in a week, the feature advantage has a half-life of about a week. Competition stops being about what you built and moves to the things that don't live in the binary. I keep landing on the same short list, and every item on it is weaker than it first looks.

Maintenance isn't it. If everyone can build the software, everyone can maintain it too. Maintenance is just more building. The same agent that cloned it waters the garden. That's not a moat, it's labor, and labor is a cost.

Audience isn't it either. You don't own an audience. You can get a Discord, you can broadcast to it, but people are people. A Discord is permission to send a message, revocable the second you bore them. It's a lease on attention, and the rent is staying good.

Simplicity is the interesting one. In a world where adding a feature is free, everyone's product bloats automatically. An agent will happily bolt on anything. It has no reason to say no. So maybe the move is the inverse of everyone else: keep the simplest product that still delivers the right value. Refuse features while your competitor ships all of them.

But be honest, simplicity isn't a moat either. Someone can out-simple you. The hard part isn't being simple, it's knowing where simple stops delivering value. That's a judgment, not a feature, and it's the one thing on the list that can't be read off a binary.

What's actually left

Strip it all back and none of them are moats. Code, features, maintenance, audience, simplicity. They're positions. You have to keep re-earning every one of them.

What's left is product sense and inertia. The sense to know which problem is worth solving, and the conviction to refuse the features that would wreck it. REA can read Photoshop's blur. It can't read why one product feels right and the clone feels dead. That part, the figuring out what to solve and the taste to know when it's wrong before anyone can prove it, doesn't copy. Everything downstream of it does.

So the moat didn't disappear. It moved. It moved from what you own to what you keep deciding.

Which leaves the real question. If features are free and copying is instant, what are you building that you'd still build knowing it'll be cloned the week it works?