What you'll learn

• 1 (00:00) **Opening Thesis: The Radical Bet** - Alejandro Maza Ayala states that Kavak is betting the company on building "superhuman agents" that outperform the best human hires on every dimension, and that the most ambitious companies will build one agent per customer.
• 2 (01:53) **The Post-Transcript Introduction** - Hosts frame the conversation: Kavak asked "What would we build if we were starting the company from scratch with AI?" rather than "How do we add AI?" They preview the discussion on tearing down a working architecture, the role of evals, and humans working for agents.
• 3 (02:06) **Alejandro's Background: Pre-Transformer ML** - Alejandro describes founding OP Analytics in 2013, a machine-learning company that was "10 years ahead of time." He explains that the ChatGPT moment made it clear they could now build a fundamentally new kind of company, which led him to join Kavak.
• 4 (03:19) **Kavak's Business and Role** - Alejandro explains that Kavak is a vertically integrated used-car marketplace (buying, refurbishing, selling, financing) that had to build fintech, logistics, and data infrastructure from scratch for Latin America.
• 5 (03:52) **The Agent Architecture: One Agent Per Customer** - The core design: when a customer arrives, a unique agent is spawned on its own virtual machine. This agent remembers years of interaction, sets a long-term goal to maximize customer lifetime value, and executes autonomously.
• 6 (05:37) **The Three Hard Decisions for the Transformation** - Alejandro details the painful, year-long process of making the architecture work, including tearing down their existing system and rebuilding it around agent capabilities.
• 7 (07:55) **Evals: The Foundation for Speed** - Alejandro explains that to move fast, you need great "brakes" (evals). Kavak spends as much engineering time and money on building evals as on building the agents themselves.

+ Full timestamped outline available in the app