BNY Mellon made an announcement in January 2026 every leader should have noticed.
Twenty thousand of their employees are now trained to build AI agents. Not use AI. Build autonomous digital workers on the bank's internal platform, called Eliza. BNY now runs 130+ "Digital Employees" independently across a 52,000-person workforce, covering over 125 live use cases.
Their tagline: "AI for everyone, everywhere and in everything."
This is not a technology story. This is a leadership story. And most leaders are not even on page one.
The Adoption Debate Is Over
If you have been waiting for AI to be "proven enough" to take seriously, the window has closed. According to Microsoft's 2026 Work Trend Index, active AI agents in Microsoft 365 grew 15 times year over year. In large enterprises, 18 times.
Your team is already using AI. Some of them are building agents. You are either ahead of this or behind it.
The real question is not "should we adopt AI?" It is: what does leading a team look like when some of the workers are algorithms?
Most leadership frameworks have no answer. Which is the problem.
Four Things Break When AI Does the Work
When your team includes AI agents executing tasks without human review at each step, four things break in your traditional leadership model.
Feedback loops. You give feedback to people. You review their work, have a conversation, watch them adjust. Agents do not grow from feedback the same way. They need to be reprompted, retrained, or rebuilt. If you do not know the difference, you will keep applying the wrong tool.
Accountability. When a human on your team makes a mistake, accountability is fairly clear. When an AI agent makes a mistake... sending the wrong report, flagging the wrong customer, misclassifying a document... who owns it? The leader who deployed it. Always. You are accountable for your agents the same way you are accountable for decisions you delegate to your team. If you deployed an agent without proper oversight, the decision sits with you.
Performance management. You cannot write an agent a performance improvement plan. You cannot motivate it. Managing an agent workforce requires a completely different toolkit. One built around outcome definition, quality standards, and governance. Leaders who try to manage agents like people are setting themselves up for expensive confusion.
Trust. When your human team sees you deploying AI to handle work they used to do, trust is at stake. Some will feel relieved. Others will feel sidelined. Most will watch how you handle it. The way you communicate intent, involve people, and create space to experiment will determine whether your human team thrives alongside AI or quietly checks out.
The Numbers Are Not Flattering
Microsoft surveyed tens of thousands of workers for their 2026 Work Trend Index. The leadership data is worth sitting with.
Only 26% of employees say their leadership is clearly aligned on AI strategy. Only 13% say they are rewarded for reinventing work with AI. These are not numbers from a technology survey. They are fingerprints of a leadership failure.
Now look at what happens when leaders do show up. When managers openly model AI use, their teams show a 30-point lift in trust in agentic AI. When managers create psychological safety around experimentation, teams are 1.4 times more likely to be active AI users who drive real results.
Organisational factors account for twice as much of the AI performance gap as individual skill does.
Read it again. The gap between high-performing AI teams and struggling ones is a leadership problem, not a skills problem.
What I See in the Field
At Step It Up HR, I work with leaders across industries on feedback systems, team performance, and accountability. The pattern I keep seeing with AI is identical to patterns I have seen with every other shift in how work gets done.
Leaders who struggle treat it as a tool problem. They buy the licences, run the workshops, then step back and wonder why nothing changes.
Leaders who succeed treat it as a people problem. They define what "good" looks like more precisely than they ever have before, because AI is unforgiving about vague expectations. They involve their teams in deciding what to automate. They stay close to the quality of outputs, not the mechanics of how work gets done.
There is a principle I come back to often: you are accountable for what you delegate, even when you delegate it to a machine.
The principle does not change because the worker is a digital agent rather than a human one. It becomes more true.
What This Means for Your Org Chart
BNY Mellon's model is instructive. They did not hire a team of AI engineers to build agents for everyone else. They trained 20,000 of their own employees, distributed and domain-specific, to create agents inside their own areas of expertise. Their CIO noted these power users "aren't in our engineering group."
The people who understand the work are best placed to define what an AI agent should do. But they need leaders who give them clarity on goals, standards, and guardrails. Without leadership alignment, you get 20,000 people building in 20,000 directions, and nobody owns the outcomes.
Look at your org chart. Some of the positions on it will, over time, be executed by AI agents. Not speculation. BNY is already there, and parts of your own organisation are already doing it whether you know it or not.
The question for leaders is not whether this will happen. It is whether you are defining how it happens, or whether it is happening to you.
Three Things to Do This Week
You do not need an enterprise AI strategy to start. You need to act like a leader who takes this seriously.
First, find out what your team is already doing. Ask directly. Where are they using AI? What agents have they built or borrowed? You will learn things worth knowing, and your team will feel seen for work they are doing quietly.
Second, define one outcome more precisely. Pick one piece of work AI touches in your team. Write down what "good" looks like in terms a machine would understand. Not "high quality." Not "timely." Specific: turnaround time, error rate, format requirements, customer satisfaction score. This is the real work of leading an AI-augmented team.
Third, remove one friction point around experimentation. Microsoft's data is clear: when leaders create psychological safety around AI, results follow. Make it clear to your team, out loud: trying something and failing is acceptable. Say it. Then repeat it when someone does.
BNY Mellon's bet is leaders of AI-augmented teams will outcompete everyone else. They are right. The advantage is not in the technology. It is in the leadership surrounding it.
What are you doing to make sure you are one of those leaders?