Every major hyperscaler now ships a production agent platform, and pilots are running at most large institutions across banking, wealth management, and insurance. The question incumbents actually face isn't which vendor to back — it's how to let a system that plans its own execution path operate safely above a core ledger, a policy administration system, or a custodial platform that was never built to be called by anything non-deterministic. Get that wrong and you've either rebuilt the estate you spent a decade hardening, or handed a reasoning model root access to a mainframe.
Join Craig Beattie, Craig Focardi, and Dan Mayo as they unpack Celent's new research on agentic architecture and legacy modernization — why the brownfield path beats a greenfield rebuild for regulated firms, and what changes once you accept that most institutions will run several agent orchestrators at once, not one.
In this session, we'll discuss:
• Why the greenfield agentic reference architecture collapses under real compliance and core-system constraints
• The brownfield alternative: treating the agent as an intelligent client above your existing API gateway and identity plane, not a replacement for either
• Practical integration patterns — API-as-a-tool, ephemeral token exchange, compensating sagas — for wiring agents safely into legacy cores
• The multi-orchestrator reality: why the goal is a shared substrate of identity, tools, and audit, not convergence on a single platform
