Human decision-making is an arduous activity requiring the complex processing of data, information, knowledge and rules for given context and constraints when choosing between two or more options. AI systems claim to provide capabilities for automating and improving human decision-making. However, AI systems and decision-automation often suffered from explainability and trust issues requiring resilience and observability by design. The challenge is: how to do so? This paper proposes the observable human-centric Agentic Neuro-symbolic (ANS) AI system architecture pattern to address this challenge. ANS AI pattern is organized into five layers: human, ANS AI system, environment, resilience and observability. Human layer interacts with the ANS AI system layer for AI powered decision-making. Based on trimodal thinking, ANS AI system contains integrated three types of AI components: intuitive, rational, and control components. ANS AI system influences or is influenced by the environment layer. Finally, resilience and observability layers are integrated into the architecture pattern for enhancing the explainability and trustworthiness of AI powered decision-making. It is anticipated that the proposed ANS AI system architecture pattern and underpinning modelling elements seem useful to support explainable and trustworthy AI powered decision-making.