Will LLMs kill software patents?

Our legal team at BigComp recently briefed us on an interesting point: if you use tools like Copilot or other LLM-based code generators in your codebase, it could make it significantly harder to acquire a software patent. At first, this seemed a bit abstract, but thinking about it, it makes sense.

The logic is that code generated by LLMs is derived from a broad corpus of existing public code, making it tricky to claim novelty—one of the key requirements for a patent. Plus, tracing the origin of generated code becomes a legal gray area.

Now that AI tools are quickly embedding themselves across the entire software development lifecycle—from design and prototyping to actual coding and testing—does this mean we’re looking at the gradual extinction of software patents? If AI-generated code can’t be patented, does that push the industry toward a future where patents play a much smaller role in software IP?

Curious if others have heard similar things or have thoughts on how this might reshape software IP law.