America's oldest bank is teaching employees to build AI agents while startups build the financial rails those agents will need to transact.
The Summary
- BNY Mellon held an internal demo day where employees pitched agentic commerce projects, signaling a shift from traditional top-down AI deployment to workforce-driven experimentation
- Para is building wallet infrastructure that gives AI agents spending power with embedded restrictions, solving the permission and overspend problem before agents scale
- The convergence point: traditional finance institutions are培养 internal agent builders while crypto infrastructure companies build the payment rails those agents will eventually need
The Signal
BNY Mellon, founded in 1784, is running internal AI demo days where employees pitch agentic commerce ideas. This is not a hackathon. This is the bank's core AI strategy, betting that the people closest to financial workflows will spot agent opportunities executives miss. The approach inverts the typical enterprise playbook where innovation happens in labs and gets rolled out to workers who had no say in building it.
The timing matters because the infrastructure for agents to actually transact is finally arriving. Para CEO Nitya Subramanian describes wallet systems that let AI agents hold and spend stablecoins autonomously, with built-in spend limits. The technical challenge is not giving an agent access to money. It is giving an agent the right amount of access, with guardrails that prevent a rogue prompt from draining accounts.
"The wallet infrastructure that lets AI agents hold, spend, and be restricted from overspending stablecoins on their own."
Para's work highlights the permission layer problem. Right now, setting spending rules for agents is manual and clunky. As agentic commerce scales, the permission structure needs to be as automated as the agents themselves. Think of it as the difference between writing a check for every purchase versus having a corporate card with category limits and real-time monitoring.
Here's what BNY Mellon's employee-driven approach gets right:
- Workers know where friction lives in financial operations
- Bottom-up experimentation surfaces use cases faster than top-down mandates
- Building culture shifts faster when people create tools for themselves, not have tools imposed on them
The gap between these two stories is the opportunity. BNY Mellon has employees dreaming up agent use cases. Para has the infrastructure to make those agents financially autonomous. What's missing is the bridge: how traditional financial institutions will integrate crypto payment rails without admitting they're integrating crypto payment rails.
The Implication
Watch for banks to quietly pilot stablecoin-based agent wallets under the banner of "digital asset experimentation" or "programmable payments." BNY Mellon's workforce strategy tells you where the demand is coming from. Para's infrastructure tells you how that demand will get met. The companies that connect employee-identified workflow problems to crypto-native financial primitives will own the agentic commerce middleware layer.
If you work in financial operations, start mapping which tasks in your workflow could be handed to an agent with a $500 daily spend limit. That is the unit of change coming.