AI adoption is no longer the interesting question. Most financial institutions are already experimenting with generative AI, copilots, Claude Code and increasingly agentic systems. The more important question is whether these systems are changing core business workflows or merely adding another interface on top of existing processes.

This distinction matters. Horizontal tools such as chatbots and copilots are useful, but they often remain productivity overlays. They help individuals draft, summarize, search, or code faster. The larger economic impact comes when AI is embedded vertically into business processes, such as investment research, Know Your Client (KYC), risk analytics, compliance review, portfolio monitoring, operations, and software delivery. In those settings, AI does not simply respond to a prompt. It observes context, reasons about next steps, uses tools, adapts to feedback, and coordinates across a workflow.

However, this is also where the real architectural challenge begins, and our mindset needs to adapt to this new challenge. Because now the problem shifts from: "Can the model solve this task?" to something closer to: "Can the system execute a stable sequence of decisions?" This is why the next phase of AI won't be defined by adding more agents, more tools, or larger models into already complex systems. It will be defined by whether firms can design agentic infrastructure that is reliable, governable, and economically sustainable.

Finance is especially well suited to this shift because many workflows are naturally decomposable. A financial analysis task, for example, can be split across agents that examine revenue trends, cost structures, market comparisons, risk signals, regulatory constraints, and client preferences before integrating the result. This is the promise of agentic AI in finance: not generic automation, but structured digital organizations that can reason across specialized subtasks.

However, this also introduces a coordination tax. Multi-agent systems are not collections of independent intelligence. They behave more like probabilistic pipelines, where every handoff can amplify uncertainty, latency, and cost. Without validation gates, observability, memory controls, budget limits, and clear orchestration logic, systems that look impressive in demos can become fragile in production. This is why topology matters. Centralized, decentralized, independent, and hierarchical agent designs distribute communication and control differently, which directly affects reliability and failure propagation.

The firms that benefit most from AI will therefore not be the ones that simply deploy the newest model first or plug Copilot into their workflow. They will be the ones that build the glue layer around AI: protocol-based interoperability, secure tool use, evaluation, monitoring, governance, feedback loops, and human oversight. Standards such as MCP and A2A point toward this future, where agents can interact with tools, systems, and each other in a more composable way.

The future of AI in finance is not just better models. It is better agent infrastructure, and carefully orchestrated multi-agent systems. This will determine whether AI remains a productivity experiment or becomes a durable source of operational and strategic advantage.

Written by Nicole Koenigstein.

Nicole Koenigstein is an AI Researcher and Practitioner in Agentic Systems, working across research, consulting, teaching, and direct system implementation to build reliable, productionready AI systems. Her work focuses on multi-agent architectures, evaluation, safety, and longterm system behavior.

She served as an external evaluator for a European Commission AI Grand Challenge and has advised IOSCO on generative AI in regulated environments. She also serves on advisory boards for leading AI and quantitative finance conferences. Nicole regularly delivers invited talks and technical workshops across academia, industry, and international events.

She is the author of Math for Machine Learning and Transformers in Action with Manning Publications. Her forthcoming books, Transformers: The Definitive Guide – Applications Beyond NLP and AI Agents: The Definitive Guide, will be published by O'Reilly Media.