On the 11th floor of BNY’s Lower Manhattan headquarters, a governance flowchart fills a curved wall of monitors: sixteen steps of review boards, compliance gates, and model risk evaluations, mapping the journey from a good idea to a working agent at one of the world’s most regulated institutions. It looks a little like a board game—but at a bank that clears US Treasury bills and moves $2.5 trillion in transactions a day, every square on the board exists so that AI is observable, controllable, and safe enough to trust.

A few dozen employees sit with laptops open; others join from offices in London, Singapore, Pittsburgh, Pune, and beyond. All of them are here to learn how to use AI to help inform decision making and solve problems. This is BNY’s AI bootcamp, a 40-hour program run during paid work hours that has graduated roughly 2,300 employees since it launched with just 35 participants in April 2025. Sessions like this one now run 800 strong.

The program’s premise is that every employee who walks out of this room should be able to design work, not just do it. By the final session, each participant is required to have built a working AI prototype—a real solution to a real problem from their own corner of the bank. “AI is changing how work gets done. We’re investing in our people to do more with AI—not merely less of the mundane, but more of everything,” says Michael Demissie, BNY’s head of applied AI.

The instinct to hand off the tedious and reclaim the valuable is happening everywhere at BNY. AI and agentic solutions increasingly absorb the work that used to grind at attention and patience; the repetitive, rules-bound must-dos that pile up around the clock, so the people alongside them can focus on what no algorithm can replicate. It’s a bold move for America’s oldest bank, and one that requires rethinking how employees work, how leaders make decisions, and how the organization learns.

Founded in 1784 by Alexander Hamilton, BNY occupies a singular position in the financial system, one of 29 institutions worldwide whose failure could destabilize markets across continents. The bank today clears most US Treasury bills issued by the federal government, holds $59.4 trillion in assets on behalf of institutions around the world, and counts 96 of the planet’s 100 largest financial firms among its clients. When BNY began exploring generative AI, the question from its leaders wasn’t “Should we?” It was “How could we possibly?”

Despite these constraints, BNY has surged ahead, investing $3.8 billion in technology last year, nearly 19 percent of its revenue, and logging 171,000 employee learning hours on AI alone. BNY has worked with Microsoft throughout its AI journey, including using Azure cloud infrastructure and placing Copilot tools in the hands of its employees. BNY is also an inaugural member of the Frontier Firm AI Initiative, a collaboration with Microsoft and the Harvard Business School AI Institute.

The bank is among the clearest examples yet of what an enterprise company on the Frontier actually looks like in practice—what it means for the people doing the work, the leaders rearchitecting it, and the institution redesigning itself. Spend a day at BNY and you will hear CEO Robin Vince’s mantra in nearly every room: “AI is for everyone, everywhere, and for everything.”

Rachel Lewis, who leads AI enablement across BNY’s operations organization, has a job that no management textbook quite covers. She oversees multiple agentic solutions alongside thousands of human staff. BNY calls these “digital employees,” and there are nearly 140 of them throughout the company: “super agents” built from multiple agents. They each have login credentials, an avatar, an employee number in the corporate directory, and a supervisor who assigns tasks and reviews output.

In the language of the Frontier Firm, Lewis is an agent boss, and her experience shows what it means for a leader to reimagine work rather than simply manage it.

For years, BNY’s payments team dealt with roughly thousands of manual interventions a day: missing country codes, malformed addresses, garbled routing fields. They tried robotic process automation. The bots had to be scripted step-by-step through every screen, and BNY’s complex infrastructure—mainframes, HTML interfaces, hybrid systems—fought those scripts constantly. “Maintaining the bots consumed so much effort that the humans were effectively doing two jobs instead of one,” Lewis says.

Last spring, her team worked with BNY engineers to build the bank’s first dedicated payment digital employee with one narrow task: read a vendor address on a cross-border transaction; call a mapping API to determine whether “York” means the United Kingdom, South Africa, or Western Australia; validate the country code against anti-money-laundering requirements; and submit the corrected payment for a human to review. The scope was deliberately tight—not because the technology couldn’t do more, but because tight scope is what makes digital employees trustworthy and auditable at a bank where a misrouted payment can freeze a billion-dollar transfer. Lewis now sees payment validations that once took five or six minutes clear in under 30 seconds. Open investigations in payments dropped by nearly 80%.

Then there’s BNY’s digital employee for client payment inquiries, which operates inside the bank’s systems similar to how an employee would. Say a funds manager in Hong Kong emails at 2 am New York time asking why a transfer hasn’t settled. The digital employee reads the message, gauges the urgency, pulls records from multiple systems, drafts a response in the appropriate tone, and queues it for review by a human manager. An inquiry that once waited days now resolves the same day.

The assumption was that managing machines would feel less human. At BNY, it’s turned out to be the opposite.

Under the direction of human managers across the organization, BNY’s digital employees are tackling reconciliation, anomaly detection, quality assurance, and more. That allows the firm to process more than 500,000 check images a day through BNY’s lockbox service, using Azure’s document intelligence to read handwriting and scanned attachments that defeated older OCR systems.

Lewis loves the efficiency but recognizes an even greater outcome: AI is beginning to change the shape of operations work itself. Tasks that once centered on repetitive exception handling are becoming more analytical, strategic, and developmental. Team members who previously spent much of their day manually repairing payment exceptions are now identifying data-quality patterns, benchmarking clients on data quality, contributing to client strategy discussions, and supporting the design and testing of new AI tools.

What does it take to be a good manager in this changing environment? It starts with “the willingness to say ‘I don’t know,’” says Michelle O’Reilly, BNY’s head of talent. “Good managers now are facilitators of dialogue. They encourage experimentation, accept ideas from everywhere, and listen for insights.” They may not know how to build the next agent. But they know how to put the right people in the room to make it happen.

O’Reilly’s team is developing what BNY calls a Digital Employee Agency—a formal process for deciding when a business problem warrants deploying digital employees rather than adding a human employee or contractor. Think of it as HR for AI: onboarding, scorecards, and eventually, when a better version comes along, a structured offboarding. “Do I need an employee? A contractor? A digital employee? Sometimes the answer’s not clear,” O’Reilly says. Her team’s goal is to remove that friction entirely: describe the work, then let the system recommend the right solution.

The assumption was that managing machines would feel less human. At BNY, it’s turned out to be the opposite. Managing digital employees looks more like running a performance-review process than operating software. Agent bosses monitor a live dashboard showing processed volume, cycle times, missed transactions, and approval outcomes. When the payments digital employee’s team noticed that some of its responses were being sent back by reviewers—not because they were wrong, but because reviewers had ingrained habits from the way the work used to be done—the team worked through it the way they would with any new hire: coaching, recalibration, retraining.

These innovations have the potential to reshape the org chart itself. BNY is preparing for a scenario in which the human workforce evolves from a pyramid to a diamond: repetitive processing work, once handled by large teams at the base, increasingly belongs to digital employees. Analytical and creative roles in the middle are expanding. Strategic leadership at the top is still making the calls that matter. BNY plans to continue adding digital employees in the months ahead, but the bank still hires thousands of human employees every year. The institution is redesigning itself around a simple conviction: that AI makes human work bigger, not smaller.

An evolving human workforce

As digital employees take on routine work, BNY sees the potential for their human workforce to evolve from a pyramid to a diamond: thinner at the ends, wider where human judgment lives.

None of this happens without serious infrastructure underneath it. BNY runs its internal AI platform, Eliza, partly on Microsoft Azure. Eliza (named for Elizabeth Schuyler Hamilton, the founder’s wife) is model agnostic by design but connects to the bank’s Microsoft 365 environment through the Graph API, drawing context from Azure, Outlook, Teams, SharePoint, and Dynamics CRM. And in November 2023, BNY became the first major global bank to make Copilot available to employees broadly. Today, tens of thousands of BNY employees—roughly half the workforce—are licensed for Microsoft 365 Copilot.

In a heavily regulated environment, having internal advocates for AI is essential. At BNY, the legal department—the group most people assumed would resist the longest—became the first to reach 100 percent in preliminary AI training. Seema Phekoo, a managing director and legal counsel leading the department’s AI transformation, helped roll out 500 Copilot licenses and custom training across legal, built agents to redline documents and reduce contract review time, and ran a hackathon with Microsoft pairing senior counsel on pro bono challenges.

“Lawyers are ideal champions for this kind of rollout,” Phekoo says. They understand guardrails. They read closely, question assumptions, and revise toward precision, which turns out to be excellent preparation for working with AI. “When a group most would stereotype as risk-averse becomes the most enthusiastic, the conversation across the rest of the firm changes fast,” she says.

Back at the bootcamp, BNY’s employees are figuring out what it actually means to work alongside AI—not as a tool you operate, but as a force that changes what you’re capable of.

A facilitator toggles a configuration setting on the Eliza interface. A checkbox next to an upload button appears. Igor Kaplevich, who works in finance on BNY’s foreign exchange desk, leans forward and starts making notes—he’d been hunting for exactly that feature. His team usually spends hours analyzing PDFs—charts, graphs, month-to-month revenue data that arrives unstructured. Eliza, being multimodal, can now read a financial chart, pull the figures, calculate the variance, and draft the explanation. “Frankly,” Kaplevich says, “this is a great way around one of the most boring parts of my job.”

Download

Frontier case study: Explore an executive overview of BNY’s AI transformation.

BNY calls it intelligence leverage: AI as a parallel to the mechanical leverage that made industrial machines indispensable in previous generations. The idea is that human work becomes more valuable, not less, when the repetitive parts run at scale around the clock. The bank is already exploring a tool that would allow any employee to record a workflow, have Eliza analyze and map it, flag inefficiencies, and then create a preliminary draft of instructions for a new digital worker. What once took weeks takes about an hour, sometimes less. AI is now helping to build AI.

What BNY is assembling, beneath the platforms and the protocols, is less a technology strategy than a theory of human potential: that people do their best work when the routine is handled for them.

Want more essential insights on AI and the future of work? Subscribe to the WorkLab newsletter