How Did Nubank Rebuild Itself Around AI?

David Vélez says Nubank rebuilt its organization around AI after it missed the December moment when the new models started writing Clojure well. The team only caught that switch about 60 days later, then cut the bank from 10 layers to five and grew his direct reports from one to 20.

Back in Brazil, the team mapped 120 workflows across the company. Coding is only one. The rest run from product design and know your customer flows to internal audit, compliance, and regulatory capital management. The 120 workflows are being reinvented natively with AI tools, with eval infrastructure built to measure every step. Some areas are further along than others. Often the huge productivity gains are not even from AI. There was always an opportunity to reinvent how work gets done, and reinventing it without AI would still find gains. AI augments that possibility, which opens a larger span of control, fewer managers, and much more visibility.

The vertical shape built up as Nubank scaled from a million customers to 140. The company took on its own inertia and started adding layers. Fast-scaling organizations see things breaking all over the place, and the easy answer when something is on fire is to add managers right away. Nubank ended up verticalized, in a structure with little visibility. AI makes it possible to step back and reinvent the business, and it was the catalyst for rethinking the structure.