Most banks racing toward agentic AI are about to learn an expensive lesson. You don’t start with the bots.

BNY didn’t. The first global systemically important bank (G-SIB) to publicly deploy what it calls “digital employees” — AI agents with logins, email addresses, and human managers — built its agentic future in the most counterintuitive order possible. Platform first. People second. Automation last.

It’s working. BNY has more than 130 digital employees in production, up from 70 less than a year ago. The bank closed out 2025 with 18% adjusted CAGR, 21% pretax income growth, and a 13% dividend bump. That’s not a vendor pitch. That’s a regulated bank executing on a strategy most of the industry is doing in reverse.

After working with BNY for months and having them to speak at Forrester’s Technology and Innovation North America in Austin last year, we just published a case study. Here’s some of what you can learn from it.

The Trust Tax Is Real, And It Will Eat Your ROI

Agentic AI looks like a shortcut. Drop in autonomous agents, let them work 24/7 to remove human workload. The fact is that most current “agentic” deployments are agentish at best: LLMs with a system prompt and a few tools, not actual autonomous workers. We are about to publish the State of Agentic AI 2026, so look out for that blog, coming soon.

Here is a catch. The moment agentish agents touch a regulated process, the trust tax kicks in. Least-privilege access. Continuous monitoring. Auditable logs of every action and every rationale. None of that is optional for a G-SIB like BNY. And none of it scales unless something underneath it scales first.

Most firms are paying the trust tax retroactively, bolted onto a platform that was never designed to carry agents and data that is not AI ready. That is why their pilots stall and their boards lose patience. BNY didn’t fall into that trap.

Step 1: Build The Platform Before You Need It

In 2023, BNY launched Eliza, a model-agnostic platform that brings Anthropic, Google, OpenAI, and other models under one governed roof. Today 97% of the bank — about 50,000 people — work on it. Eliza has powered more than 160 production AI solutions, a 200%+ year-over-year jump.

That isn’t a side project. That is the foundation that made every later step cheap and fast. Discoverable use cases. Reusable AI assets. Dashboards that surface governance artifacts and connected data. When the trust tax hit, the platform absorbed it.

Step 2: Train The Workforce Like You Mean It

Then BNY did the part most companies are just now wading into. It taught its employees how to use what it built.

More than 1,400 employees graduated from 40-hour live bootcamps in 2025. The bank delivered over 170,000 hours of AI learning across bootcamps and self-service paths. Today, more than one in three employees has built a custom agent on Eliza — summarizing risk reports, generating client-ready insights, automating the actual work they own.

Workforce readiness is the part that gets cut from most AI strategies and shouldn’t. As Michael Demissie, BNY’s head of applied AI, put it: “The bigger challenge isn’t whether the technology works — it already does. The challenge is getting people to embrace it.”

Bottom-up adoption is the multiplier. You can’t buy it. And your employees won’t embrace it if they perceive it as a threat to their jobs.

Step 3: Now Launch Digital Employees

Only after the platform was live and the workforce was fluent did BNY start deploying real autonomous agents in core operations. Payment validation agents that identify issues with stuck transactions in minutes instead of hours, finding problems that humans miss. Engineering agents that repair code. Each one scoped, monitored, and fully auditable.

Demissie again: “Every action the digital employee performs is fully auditable… the rationale is logged.” That is what agentic AI looks like when the foundation is real.

The Lesson For The Rest Of Us

If your agentic strategy starts with the agents, you are already behind. The fastest path to scaled, trusted, ROI-positive agentic AI runs through a governed platform and a fluent workforce — in that order. BNY did the unglamorous work first. That is exactly why its digital employees are real while most of the industry is still arguing about definitions.

Read the full case study, Case Study: BNY’s AI And Agentic Strategy Is Paying Off, then schedule an inquiry. We’ll help you sequence your own agentic strategy before the trust tax catches up with you.