Do I own the code, the AI models, and all the IP when the project is done?

Why IP ownership isn't automatic in the U.S.

U.S. copyright law grants initial ownership of creative work — including software — to the person or entity that created it. Unlike a direct employee (where "work made for hire" doctrine often applies automatically), a contractor or dev shop retains copyright unless a written agreement explicitly transfers it to you. This is a common, expensive surprise for founders and operators who assumed the code they paid for was theirs.

AI-generated outputs add another layer: model weights, training pipelines, and fine-tuned models each raise their own questions about ownership and licensing. A contract that covers source code but is silent on AI artifacts can leave critical assets in a gray area.

What a proper IP assignment should cover

• Source code — all custom-written code, scripts, and configuration files.
• AI models and weights — any fine-tuned, custom-trained, or distilled models built specifically for your product.
• Training data and datasets — proprietary data pipelines and labeled datasets your project created.
• Designs and documentation — UI/UX assets, architecture docs, API specs.
• Inventions and derivative works — anything derived from the above during the engagement.

The agreement should use the phrase "assignment of all right, title, and interest" — not merely "license to use." A license lets you use the work; an assignment means you own it.

Third-party components and open-source

Almost every modern software project includes open-source libraries, third-party APIs, or foundation AI models (e.g., open-weight LLMs). These carry their own licenses — MIT, Apache 2.0, GPL, and others — that your dev partner cannot assign to you because they don't own them either. A reputable partner will disclose all third-party dependencies and confirm they've used components whose licenses are compatible with commercial ownership. Ask for a software bill of materials (SBOM) at project close.

Red flags to watch for

• Contracts that grant you a "perpetual license" rather than full ownership.
• Vague language like "work product" without defining what it includes.
• No mention of AI-generated artifacts, model weights, or training data.
• Clauses that let the vendor reuse your code or models for other clients.
• No NDA in place before discovery or design work begins.

How CodeNicely handles this

CodeNicely operates on an NDA-first, full IP assignment model: clients own 100% of the code, AI models, and all related deliverables at project close — no vendor lock-in, no reuse of your IP for other clients. Milestone-based delivery means you receive and can verify assets incrementally, not just at the end. If you're evaluating any development partner, ask them to point to the specific clause in their standard contract that assigns IP. If they can't, that's your answer.