Summary
Ali Masa, Head of AI at Kavak — a Latin American used car marketplace that has expanded into fintech, logistics, and financial services — describes how the company rebuilt itself around AI agents in an a16z podcast interview. Kavak’s transformation involved significant organizational downsizing and nearly a year of failed attempts before arriving at a working architecture, and Masa is candid about both the difficulty and the stakes involved.
At the core of Kavak’s system is an approach that deliberately moves away from workflow-based multi-agent designs. When a customer interacts with Kavak, an agent is instantiated specifically for that customer, running in its own virtual machine with access to years of prior interaction history. The agent sets long-term goals around customer lifetime value and pursues them autonomously — approving car loans in under three minutes, managing complex financing decisions with deep personalization, and adapting across multiple financial products. Between 100,000 and 200,000 such agents are instantiated every day. Kavak also launched a “Jedi Academy” program that trains all employees — from mechanics to the CEO — to build and ship production AI agents within six weeks.
The most striking experiment Masa describes is deploying an AI agent as the CEO of an entire city — Guanajuato, Mexico — with a one-month goal of doubling profits. The agent achieved a 1.5x increase. The conversation also covers the design of evals for high-stakes financial decisions, the difficulty of building reliable long-running agents, and the philosophical question of how to design AI systems when human-level leadership roles are now plausibly within reach.
📺 Source: a16z · Published August 10, 2026
🏷️ Format: Workflow Case Study
