Patent attorney Yongduck Kim's column, “Patenting Agentic AI: What Regie.ai Reveals about the Next Stage of Enterprise AI Competition,” examines how a patent defines agentic AI as an invention.
Regie.ai's U.S. Patent No. 12,346,822 B2 expresses an agent as a system of states, actions, constraints, rewards and retraining, illustrating a shift in enterprise AI competition.
The column starts with Regie.ai's October 7, 2025 announcement and examines autonomous action selection through the patent's claims and components. It describes a decision engine that allocates resources under organizational constraints and learns from execution results, beyond generating sales emails.
Protecting a decision engine
The most obvious part of this patent is not “automation” but “decisions.” The system takes input from multiple data sources, making the current state calculated, and then chooses the next action based on that state. In contrast to the way people run workflows by manually tagging things like interest and non-response, the start is different in that the state itself is a dynamically generated representation.
Claiming action selection
A reinforcement-learning model selects actions that encompass channels, timing and contact methods, beyond a single instruction such as sending an email. The patent incorporates a constraint model into that selection. Even where conversion prospects are strong, staffing, time, cost, regulatory requirements and customer fatigue limit the available choices. The architecture determines how actions are allocated under those resource constraints.
Closed loops leading to creation, execution, and relearning
A generative model converts the selected action into executable content, such as an email or call script, and passes it to an external system. Outcomes such as responses, meetings and revenue contributions feed back into learning. This creates a closed loop that evaluates performance and updates the policy.
The Next Battleground of Corporate AI
The column argues that enterprise AI competition is expanding from content generation to reliable decision-making under organizational constraints. Patents on these decision architectures may require competitors to consider claim scope and design-around strategies as well as implementation.
IPLEX IP Law Firm supports patent strategy based on technology and business structure in advanced technologies such as artificial intelligence, smart factory, and blockchain. Yongduck Kim patent attorney has worked on intellectual property rights of leading domestic and foreign companies such as LG Electronics, Samsung Electronics, Su-A Lab, MakinaRocks, and Xiaomi. He has accumulated expertise in various fields such as the evaluation of outstanding products related to artificial intelligence and IoT technology, and professional evaluation of technology-based special listing.
IPLEX will continue combining technical analysis and IP strategy to help businesses protect and commercialize innovations in enterprise AI.
This article reflects the information available when it was published. Contact us to discuss your circumstances.