Interview, Fireside Chat
Kavak's Playbook for Rebuilding a Company Around AI
a16zAngela Strange, Gabriel Vasquez, Alejandro Maza Ayala, Ale Massa, Carlos, Gabe, Joseph Schumpeter, Edison, Ford
- Ale Massa, Head of AI at Kavak, bet the company on transforming into an AI-native organization by building "superhuman agents" designed to outperform the best human hires on every dimension, including conversion rates, lifetime value, and customer experience.
- Kavak's current architecture deploys a dedicated AI agent per customer, instantiated daily with its own virtual machine to remember years of interaction history, set long-term goals, and maximize customer lifetime value.
- The company reports that between 100,000 and 200,000 agents are instantiated every day, with individual agents running tasks ranging from three minutes to three days before returning to a dormant state.
- Approximately 96% of all customer interactions and 95% of all transactions at Kavak are now handled entirely by AI agents without human intervention.
- The initiative required a fundamental shift from measuring transactional metrics (cars sold) to relational metrics focused on maximizing the lifetime value of 10 million customers across various high-ticket products.
- Sales agents outperformed human teams by converting 50% more initially, now achieving a conversion rate 2.1x higher than the previous best human sellers.
- Customer satisfaction metrics (NPS) tripled after replacing human customer service with autonomous sales agents capable of handling complex financing, insurance, and trade-in processes.
- Loan approval times for customers were reduced from two months to under three minutes through AI-driven underwriting and risk assessment.
- Warranty costs for mechanics decreased by 26% after the introduction of "El Miko," an AI sidekick that assists human mechanics with inspection tips and repair procedures.
- An experiment deploying an AI "CEO" agent in Cuernavaca, Mexico, resulted in a 1.5x increase in city-level profits within the first month by micromanaging daily operations and forecasts.
- Ale Massa revealed a strategic pivot where the company destroyed two years of existing multi-agent workflow infrastructure to adopt a new paradigm of single, long-running agents with recursive self-improvement capabilities.
- The new architectural paradigm leverages agents with access to memory, evals, and a CLI to interact with every API, prioritizing long-term goals over rigid workflow graphs to maximize intelligence scalability.
- Kavak implemented the "Jedi Academy," a six-week internal training program required for all employees, from the CEO to mechanics, to learn how to build, collaborate with, and optimize AI agents.
- The organizational structure has shifted to flat, senior-led teams where humans either build agents, work under agents, or perform physical tasks, effectively eliminating traditional middle management layers.
- Ale Massa advises future founders to pursue top-down transformation strategies with a clear five-year vision, rather than relying on bottom-up hackathons, to successfully restructure companies around AI.
- A specific evaluation framework for AI spending is proposed, categorizing tokens into three tiers: Tier 1 (indirect value like chatbot usage), Tier 2 (indirect codebase value), and Tier 3 (direct ROI from autonomous agents).
- The speaker cites Joseph Schumpeter's concept of "creative destruction," arguing that significant value will be generated by new AI-native companies rebuilding from scratch rather than incumbents attempting superficial AI adoption.
- Ale Massa predicts that by 2035, with GPT-10 level intelligence, AI will likely be capable of performing the CEO role entirely, reinforcing the need for organizations to be designed for self-improvement.
- The strategy treats the organization itself as a self-improving loop, harnessing new models to deliver exponential economic value growth rather than just linear intelligence gains.