AI Scales. Lunch Doesn't
On fish, trust, and why learning is relational
Capacity development is a technical term for a simple idea: helping people get better at hard jobs. In my world, that means working with senior central bank officials in emerging economies — people who have to make consequential decisions about how money moves, how data gets classified, how policy holds up under pressure. The goal is not just to teach them a framework. It’s to give them the judgment to use it when things get complicated, and the confidence to know they are not alone when they do.
After every trip, I come back to Washington and write a report. But not everything I experience makes it into a report. Over years, that report becomes the basis of the roadmap for the next phase of programs.
The Case for Efficiency
Across the institutions that do this work including the multilaterals, the development banks, the technical assistance providers, the conversation has shifted. Not whether to use AI. How fast and how much.
The logic is straightforward. One person. Flights halfway around the world. Hotels, ground transport, weeks away from headquarters. Do the math.
An LLM can deliver the same methodology, answer questions at any hour, remember every prior conversation, and reach ten countries simultaneously. No per diem required. For organizations trying to build capacity at scale in countries that need it most, that is not an unreasonable proposition.
I have given the same lecture more times than I can count. The frameworks travel. A well-designed AI tool could reach a hundred policy maker instead of twenty, and do it faster than any team of experts ever could.
But the lecture is never really the point.
The Woman Who Waited at the Door
I noticed her immediately. She jumped into every discussion, the kind of person who makes a room feel more alive just by being in it. A participant in one of our courses in Singapore. Months later I was on a mission to her home country. She was waiting at the entrance of the central bank when I arrived and hugged me like we were old friends. In a city I barely knew, that meant something.
We spent the days doing necessary, grinding work which mostly includes reviewing balance sheets, checking classification methodologies, sitting through long institutional meetings. But one afternoon she took me to lunch at a local place, the kind that doesn’t make any list but that locals know. She ordered fish. It reminded me of home.
Eighteen years at that central bank. When I asked why, not skeptically, just curious — she said: as a child, from a rural agricultural family where no one had gone to university, she had traveled with a neighbour to the capital city and opened a bank account. She saw the bankers. How they dressed, how they read, how they worked with precision and purpose. That image stayed with her. It became the reason she pursued an education, went to banking school, and now leads international work on balance of payments for her country’s central bank.
A bank account. A glimpse across a counter. Eighteen years of work.
Will AI Have This Lunch Over Fish
Real capacity building outlasts the training room. It’s the email two years later that starts: “I remembered something you said.” The confidence to hold a position when the data contradicts the established theory comes from knowing there is a person you can call. Someone who sat across from you and meant it.
Would an AI module have reached her in that Singapore classroom? Probably. But would it have sat across a table from her, over fish that tasted like home, and heard what she had never put in a report?
And would she have told it?
Maybe, eventually.An LLM is available at any hour, across any time zone. It won’t forget what you said last time. That kind of consistency might start to feel like trust. Maybe the trusted advisor of the future won’t need a boarding pass—or even a phone call.
In some ways, that line is already blurring. We advise over Teams. We build working relationships over email. But not everything has moved.
What I do know is this: I’ve carried that afternoon into every efficiency meeting.
I didn’t write it in my back-to-office report.
Maybe I should have.