BNY Mellon’s AI Agent Push: 400 Employees, 200 Deployments, One Platform

BNY Mellon, the 240-year-old custodian bank holding roughly $50 trillion in assets, gathered nearly 400 employees for an internal AI demo day on September 6, 2026. The event showcased pitches built on Eliza, the bank's proprietary platform for creating and deploying AI agents. With 100 to 200 AI solutions already live internally and Google Cloud's Gemini Enterprise powering the stack, BNY's move signals that enterprise AI has shifted from pilot programs to institution-wide workforce transformation.

What happened

Detail	Fact
Demo day attendees	Nearly 400 employees
Assets under custody	Roughly $50 trillion
AI solutions deployed internally	100 to 200
AI platform	Eliza (proprietary)
Cloud AI integrated	Google Cloud Gemini Enterprise
Bank age	240 years

BNY Mellon ran an internal demo day where employees pitched AI agent concepts to colleagues. The format is deliberately bottom-up: frontline staff identify problems in their own workflows, build agents to fix them, and present results in an open forum. A central AI team does not hand down finished products.

The bank has built a governance framework around Eliza to keep creativity within compliance boundaries, which matters a great deal for a custodian bank operating under financial regulators.

What Eliza actually does

Eliza is BNY’s in-house platform for building, testing, and deploying AI agents. It is designed so employees without computer science backgrounds can create agents through bootcamp-style training. According to BNY, thousands of employees have now completed that training.

The bank integrated Google Cloud’s Gemini Enterprise into Eliza to strengthen its research automation and data synthesis capabilities. That integration positions Google Cloud as infrastructure, not just tooling, inside one of the world’s largest financial institutions.

Agent Maven: the standout demo

One tool highlighted at a NY Tech Week event was Agent Maven, an agentic automation tool with two specific jobs:

• Trade settlement monitoring: The agent watches for potential settlement failures before they become expensive problems, acting proactively rather than waiting for a failure report.
• Sales lead qualification: It helps relationship managers focus on the prospects most likely to convert, cutting time spent on low-probability outreach.

Both use cases share the same logic: remove a high-volume, repetitive judgment call from a human’s plate and hand it to an agent that runs continuously.

Why it matters

BNY’s approach shows that the strategic question for large enterprises is no longer “should we use AI” but “how do we turn existing staff into AI builders.” Top-down deployments, where a specialist team builds tools and then pushes them to the business, are slower and produce more integration friction. BNY’s model distributes that capability across thousands of domain experts who already understand the problems.

The Google Cloud partnership is also a signal worth noting. Major cloud providers are actively competing to sit at the AI infrastructure layer for traditional finance. Winning a deal inside a $50 trillion custodian bank is a significant reference point for any competing cloud vendor.

For businesses outside finance, the lesson is the same one we see across industries: the organizations moving fastest are the ones investing in training existing employees to build and operate AI tools, not just licensing software and hoping for adoption. This pattern is exactly what platforms like workflow automation services are designed to support at smaller scale.

Our take

The demo day format is clever governance. It creates peer pressure to actually ship something, surfaces useful ideas organically, and builds internal credibility for the AI program without requiring a mandate from the top. It also means BNY gets hundreds of domain-specific problem statements for free, because employees will only pitch agents that solve real pain in their daily work.

The 100 to 200 deployed solutions number is worth scrutinizing. “Deployed” in enterprise AI can mean anything from a production system handling millions of transactions to a Slack bot one team uses occasionally. BNY has not published performance data for these agents, so the scale of actual business impact remains unclear.

That said, graduating thousands of employees through AI builder training is a structural change, not a marketing claim. If even a fraction of those employees ship one useful agent each, the compounding effect on operational efficiency is real. We cover similar workforce-plus-AI dynamics across industries in our AI and automation news coverage.

For any business owner watching this: if one of the most regulated, risk-averse institutions in the world is training frontline staff to build agents and running internal demo days, the “it’s too risky for our business” objection to AI adoption gets harder to sustain.

What to do about it

1. Audit your own repetitive workflows. BNY’s most useful agents target high-volume judgment calls like settlement monitoring and lead scoring. Find your equivalent.
2. Start with one department, not a company-wide rollout. BNY’s bottom-up model works because small teams own their agents. Pick one team, one problem.
3. Set governance rules before you scale. BNY built compliance guardrails into Eliza from the start. Decide upfront what an agent is and is not allowed to do autonomously.
4. Consider whether AI integration support makes sense before investing in building internal training programs from scratch.

The practical takeaway: the competitive gap in AI adoption is no longer about access to models, it is about how fast you can train your existing team to use them.

Source: Bing News · Make.com