The increasing use of artificial intelligence in financial systems is driving the industry towards a model where machines execute transactions at scale, presenting new challenges related to control, oversight, and infrastructure, according to executives from Microsoft and Chainalysis. Bill Borden, Microsoft's corporate vice president of worldwide financial services, stated that legacy systems will face growing pressure as transaction demands become more complex, with the tipping point occurring when latency, scale, and complexity impact a firm's ability to compete.

This forces companies to reevaluate how their systems are built. While automation has long been a part of finance, the focus is now shifting from capability to trust, with the question being whether technology can be trusted to execute tasks such as hedging strategies, and whether it can be audited and controlled. To manage this transition, Microsoft is developing tools, including systems that assign identities and permissions to AI agents and track their actions, which is essential in regulated environments where firms must demonstrate what controlled a system and whether it followed policy when decisions are made without human input.

Jonathan Levin, co-founder and CEO of Chainalysis, noted that the crypto sector already offers a working model of automated finance, with blockchain networks processing large volumes of transactions through smart contracts and software-driven wallets, creating an environment similar to agent-based systems. Levin pointed to efforts to track illicit funds across thousands of wallets as an example of the kind of monitoring needed in a system where transactions occur at scale without human input. Looking ahead, both executives expect a mix of systems to coexist, with Levin predicting that the majority of commerce will be settled on public infrastructure in 10 years, while Borden envisions a more integrated approach linking public blockchains, private networks, and existing rails, with traditional rails continuing to exist and software acting as the connecting layer.