PromptCraft’s Gray Zone: Owning Agentic AI Workflows Without Owning Ideas
Abstract
As organizations operationalize AI, competitive advantage increasingly resides in “promptcraft”: curated prompts, agentic workflows, system instructions, and the evaluation infrastructure that reliably elicits high-value outputs and actions from models and agents. Yet promptcraft sits uneasily within existing intellectual property categories. Copyright may protect expressive prompt text, but many prompts function as methods (“procedural instructions aimed at producing a result”) implicating idea–expression limits and merger doctrine. Patent law offers limited help: prompt techniques evolve rapidly and often lack the novelty or disclosure incentives traditional inventions require. In practice, firms default to trade secret and contract, but these tools face distinctive frictions: prompt leakage through outputs, reverse engineering via repeated querying, employee mobility disputes, and difficulties proving misappropriation when independent development converges on similar solutions. This paper proposes a functional taxonomy of promptcraft—expressive, operational, and hybrid assets—and maps each category to plausible legal protections and failure modes across copyright, trade secret, unfair competition, and related regimes. It argues for a calibrated, layered approach: treat operational promptcraft primarily as protectable know-how under trade secret and contract, reserve copyright for genuinely creative prompt expression, and deploy unfair competition doctrine to address appropriation that neither regime captures. The result is a practical framework for ownership, enforcement, and sharing in AI-enabled production.
Copyright & permissions
