Agentic AI offers potential to streamline central bank policy briefing
Agentic AI, combining large language models with autonomous goal setting, can automate key tasks in central bank policy briefing. This approach aims to reduce the administrative and analytical burdens imposed by the increasing complexity and frequency of policy documents.
Automating central bank policy workflows with AI
Central banks increasingly rely on vast volumes of policy documents, ranging from monetary policy statements to regulatory guidelines and internal reports.
The complexity and frequency of these documents impose significant administrative and analytical burdens.
This paper explores how agentic AI, which combines large language models with autonomous goal setting, structured workflows, and tool-calling capabilities, can automate key tasks.
These include drafting, summarization, cross-document consistency checks, integration with data sources, and the generation of tables and figures.
The study presents two practical examples using open-source tools, demonstrating how development and operating costs can be controlled.
It concludes by discussing how agentic AI could reshape document processing within central banks, emphasizing both efficiency gains and the necessary governance safeguards for trustworthy adoption.
From passive LLMs to autonomous agents
The development of AI has advanced significantly from basic large language models (LLMs) to autonomous agents.
Early LLMs, while revolutionary for text generation, were limited by static data and occasional inaccuracies.
Retrieval-Augmented Generation (RAG) addressed this by integrating real-time information, enhancing factual grounding.
Further advancements focused on instruction following and alignment, making LLMs more reliable and conversational through fine-tuning.
A crucial step involved integrating external tools and plugin architectures, enabling LLMs to interact with databases and APIs.
This progression culminated in agentic AI systems, which embed reasoning and action in iterative loops, allowing them to decompose complex tasks, execute external functions, and refine outputs autonomously.
Efficiency gains require governance
The paper convincingly demonstrates the significant efficiency gains agentic AI can bring to central bank operations.
However, the emphasis on robust governance safeguards underscores the critical challenges in ensuring trustworthy adoption.
Without clear frameworks for accountability and oversight, the promised benefits risk being overshadowed by unforeseen operational and reputational complexities.