There is an increasing potential for using AI applications in finance, ranging from simpler Generative AI applications to more complex, agent oriented solutions. This paper reports on our experience in applying early AI solutions in an Australian fintech landscape. We first present a framework developed to support industry experts and practitioners in adopting AI solutions in a scaleable manner, to ensure the adoption of fit-for-purpose AI systems. We then focus on a longer term research dimension, which addresses more complex business problems for which the emerging multi-agent AI technologies may offer more value. We experimented with these technologies, including their integration with more mature approaches such as RAG. Our proof of concept for retirement planning application, highlights benefits and directions for LLM-powered AI agents, and also identifies limitations of current technologies. Specifically, deploying multi-agent technologies on low-powered infrastructure presents challenges. These limitations can hinder the implementation of solutions that require reliable reasoning and collaboration. Our proof of concept highlights both the potential of multi-agent technologies, and the limitations that need to be addressed.