America's Oldest Bank Is Building Its Own AI

A company builds its own AI platform the way BNY did: it builds the platform first, enables every employee second, and attacks priority use cases third, in that order. I watched the team from BNY walk a Pittsburgh audience through exactly that sequence this week at AI Horizons, and the discipline of it is the whole lesson. The oldest bank in America, founded by Alexander Hamilton in 1784, is not renting its way into this era. It is building.

That should stop you for a second. Banking is one of the most heavily regulated businesses on earth. BNY sits on more than two centuries of institutional caution and a balance sheet that measures custody assets in the tens of trillions. If any company had a reason to wait for a vendor to package AI into something safe and familiar, it is this one. Instead, BNY chose to own the capability. The question I get most often from small and midsize business owners is whether they should build their own AI or just buy it. BNY's answer, demonstrated rather than argued, is that you do both, and the order matters.

What is BNY's Eliza, and why does it matter?

Eliza is BNY's own enterprise AI platform, a company-owned layer that brings frontier and open-weight models together with the bank's internal data and compliance controls¹. The name is a tribute to Elizabeth Schuyler Hamilton, the wife of the bank's founder¹. That detail signals that this is treated as part of the institution's identity, not a tool bolted on for a quarter.

Eliza was built to hold many models at once, open source and commercial alike, rather than committing the bank to a single provider¹⁴. The models run inside BNY's own environment, governed by the bank's own rules. When a better model appears, BNY can bring it in without rebuilding its work or handing over its data. The platform is the asset. The models are ingredients the platform can swap.

That is a working example of a principle I have written about as the Intelligence Estate, and of the rule I call owning the center and renting the tails. You own the platform, the data, and the workflows that make your business what it is. You bring in models as you need them. BNY did not choose between build and buy. It built the part that compounds and bought the part that commoditizes.

Why would a 240-year-old bank build instead of only buying AI?

Because renting alone would have meant handing its most valuable material, its proprietary data and its regulated workflows, to a platform it did not control. A bank cannot outsource the thing that makes it a bank. Every fine-tuned behavior, every workflow an agent learns, every institutional judgment encoded into a process is either an asset BNY keeps or an asset that lives inside someone else's product. BNY chose to keep it.

This is the build-versus-buy question answered by a company that had every incentive to take the easy path. Buying still happens everywhere inside Eliza. BNY buys and rents models constantly. What it refused to rent was the platform, the data layer, and the accumulated capability. In my experience that is the right line to draw, and it is the same line a much smaller company can draw at its own scale.

BNY chose to own its destiny with AI rather than let a vendor own it for them.

How did BNY actually do it? The three phases.

BNY built its capability in three deliberate phases, and the sequence is the part worth copying. The team was clear that this has been a multiyear effort, not a launch. Rushing any phase would have undercut the ones that followed.

Phase one was the platform. BNY built Eliza first, as model-agnostic infrastructure on its own hardware, so that everything after it would have a governed place to live. Get this wrong and every later use case becomes a one-off integration with its own security review. Get it right and each new solution inherits the controls already in place.

Phase two was full enablement, and this is where most companies would have flinched. BNY committed to AI culturally by requiring nearly its entire workforce to build a foundation, with almost all employees completing a training program and thousands going further through a multi-day bootcamp that helps non-engineers automate parts of their own jobs²³. This is the opposite of leaving AI fluency to individual initiative, which I have argued is a strategic liability rather than a training gap. BNY treated understanding as something the whole company builds together, in coordinated fashion, so that the people adopting the technology were the same people who helped shape it.

Phase three is the one they are in now: attacking priority use cases. With the platform in place and the workforce fluent, BNY is releasing solutions against its highest-value problems and adjusting its processes as it goes. The team said they now run roughly 160 digital employees, which are AI agents given identities, access controls, and defined workflows so they can work alongside people as teammates. Public reporting from earlier this year put the count of live solutions north of 125 and climbing, which tells you the direction of travel². The BNY team also described relationships that keep them near the front of the field, including work with Carnegie Mellon and early access to Anthropic's Mythos model⁵. I could not independently confirm those last details, so I pass them along as the bank described them from the stage.

Do small and midsize businesses need to be on the cutting edge like BNY?

No, and that is the most important thing to take from this. A small or midsize business does not need to sit on the cutting edge, run its own model training, or spend a fraction of what a global bank spends. What it needs is the same posture and the same sequence, sized to its reality. BNY blazed a trail here at enterprise scale and under regulatory pressure most companies will never face. The trail is what you follow. You do not have to walk it in the same boots.

Proxigee Services helps small and midsize businesses take their first strategic steps into owning AI, choosing the right use cases, building the right foundation, and developing the internal capability to grow it. Translated to that scale, BNY's three phases become very ordinary decisions. Start with a foundation that is not welded to one vendor, so you can adopt better models as they arrive. Enable your people together rather than hoping a few enthusiasts figure it out alone. Then point the capability at one problem that actually pays, ship it, and let what you learn shape the next one.

The barriers to building have never been lower. Capable open models are free to download. A machine that runs a serious model now fits on a shelf and costs about what a company laptop refresh does. The gap between BNY and a fifty-person company is a matter of scale, and no longer a matter of access.

BNY had every reason to wait and chose to build. In the wake of a company like that, a smaller business has fewer excuses and more room to move. Own the center. Rent the tails. Start the part that compounds while the window is still open.