Presentation: Decision Models in Agentic Architectures: From Production to Agent Skills

During the last 12 months, the world of IT has undergone significant changes, particularly in the realm of artificial intelligence (AI). Early on, AI was primarily used for summarization, content writing, script, test, and configuration file generation. However, as models evolved, it became evident that AI could revolutionize backend code and critical systems for clients.

Alex Porcelli, Co-founder and CEO of Aletyx, reflects on his own journey with AI. After 30 years of software development, he began interacting with AI agents differently, relying on them to assist in his work. Porcelli emphasizes the importance of sharing experiences from the AI community to help others navigate this new landscape.

Porcelli highlights the challenge of non-determinism in AI decisions, a major issue for enterprise AI. While AI models can produce varying outputs for the same input, this inconsistency poses a problem for high-stakes decision-making processes. To address this challenge, Porcelli proposes integrating Decision Model and Notation (DMN) decision models with large language models (LLMs) and agent skills, creating auditable, deterministic agentic architectures.

By combining DMN decision models with LLMs and agent skills, Porcelli believes that businesses can maintain control over decision logic while ensuring robust architectural governance. This approach allows for consistent and accountable decision-making, which is crucial in regulated industries where explanations of decision-making processes are required.

By integrating these elements, Porcelli aims to bring domain knowledge to the age of AI, enabling businesses to leverage AI effectively while maintaining accountability and consistency.