Before bundling AI-written open-source code, run a three-question check

You are staring at a repo that looks like the missing component of your product. It has a clean UI, a README with screenshots, and a license badge that looks promising. Copying the code is the easy part. The hard part is deciding whether the next stretch of work will be spent fixing what the model left unfinished.

A recent project shows the trap. Brandon Thomas, an Atlanta-based developer, said he used Claude Opus 5.5, an AI model, and Rust, a systems programming language, to recreate seven Adobe applications, including Photoshop and Illustrator. The tools were Rust clean-room replacements for Adobe's closed programs, with browser builds included. PhotoCraft reported full menu coverage, with pixel-matching accuracy near 65%. Completion levels were uneven: EffectCraft was stated at about 94%, FilmCraft at about 85%, and PrintCraft at about 35% without a PDF renderer.

For a solo operator, bundling a recreated creative tool is a maintenance decision first. You are adopting a codebase whose author may be an agent, a solo builder, or both. Spend one hour before build so the gap is visible before launch.

Answer these three questions before you copy the code

Each question below is a requirement, not a score. If you cannot answer it cleanly, stop and treat the repo as a demo, not a dependency.

• Parity: Does the code actually do the job you need, or only the job that looks impressive in a demo? Run the core workflow on real input, not sample files. If the output is close enough for your customer, note where it falls short.
• Missing core output: What is the core output your customer expects that the project has not shipped? A missing renderer, exporter, or format can turn a small omission into a whole product.
• License: What does the license allow you to do with the code, and what does it require you to publish? Read the license file, not the badge.

Parity matters because a demo can hide what a menu cannot show: a working pipeline. Missing output is often the output your customer pays for. The license comes before copying because a permissive badge can still carry obligations you did not expect.

License is the last question, not the first

A license badge is only a summary; the contract is the file. The seven repositories carried MIT or Apache-2.0 licenses, and they ran natively on macOS, Windows, and Linux, plus WebAssembly. Apache 2.0 allows commercial use, modification, and redistribution without extra limits. A solo builder can often bundle that code without immediately triggering a release obligation. But the GPL is different: it requires anyone who publicly uses a modified version to publish the modified source code.

Joseph Saveri Law Firm and Matthew Butterick filed a late-2022 class action alleging Microsoft and OpenAI violated open-source licenses in connection with GitHub Copilot. You do not need to solve that legal question before you start. You do need to know whether the code you are bundling came from a source whose license you can actually read and defend.

Read the bug reports as your maintenance estimate

The bug tracker is the most honest part of the repo. Few issues mean the maintenance estimate is unknown, not absent. Mostly cosmetic issues may be something you can ship around. Issues clustered around export, performance, or file compatibility are the list of things you will fix after launch.

Community threads tell you the same thing from the outside. The project hit the Hacker News front page, drawing 107 points and 165 comments. That kind of attention is useful, but it is not a quality certificate. A front-page thread can be full of excitement, and the excitement can be about the idea, not the code. Count the hours you would spend answering the questions in those threads, because those are the hours your customers will ask you to spend.

The developer said the free MIT/Apache-licensed apps would be supported by selling token-based access to Artcraft's current visual AI model and IDE. That business model is not yours. It matters because it tells you what the author is optimizing for. If the free repo exists to lead users to a paid model, the missing features may be missing on purpose. Read the issue list before you decide whether that gap is yours to build.