From Copilots to Agents: Rebuilding the Company Around AI

Episode Summary

Executive Summary: Carlos Garcia Otati explains how Kavak became a vertically integrated car platform across Latin America and the Middle East by solving fraud, financing, warranties, and logistics in used cars. He details his upbringing, why co-pilots failed, and how Kavak shifted to AI agents, accepting a year of flat growth to redesign the business for the post-ChatGPT era.

Main Topics: Kavak’s all-in-one car marketplace model (Priority: 5/5): Carlos describes Kavak as an end-to-end platform for selling, buying, financing, servicing, warranting, and even handling tickets for car owners, aiming to manage the full lifetime of the user’s vehicle ownership. Upbringing, mobility, and founder mentality (Priority: 4/5): His childhood in Venezuela, frequent school changes, and family instability shaped his comfort with change, empathy across socioeconomic groups, and ability to read people and build win-win relationships. The Latin American used-car problem (Priority: 5/5): Carlos explains why the market needed Kavak: extreme informality, fraud, low financing penetration, and unsafe peer-to-peer transactions make car ownership hard to access and risky in the region. Building infrastructure businesses beneath the product (Priority: 5/5): Kavak needed to create multiple businesses beneath one consumer experience, including e-commerce, reconditioning, financing, and logistics, because Latin American markets lack the infrastructure assumed in the U.S. Early conviction in AI and system-wide data ontology (Priority: 5/5): Carlos says the company anticipated AI by building a shared data ontology and ‘robot pipeline’ early, so systems could communicate and manage edge cases at scale before ChatGPT made it mainstream. AI transition, org redesign, and adoption failure of co-pilots (Priority: 5/5): Kavak first tried employee-facing AI tools but found adoption poor, so it pivoted to agents embedded directly in critical workflows and customer interactions, eventually handling most interactions with AI. Founder discipline, self-introspection, and long-term thinking (Priority: 4/5): Carlos describes annually ‘firing himself’ to reassess whether he is still the right CEO for the phase of the company, emphasizing humility, adaptation, and the compounding value of small improvements.

Key Arguments: Latin American used-car markets are structurally broken due to fraud, informality, and weak financing; solving these frictions creates real consumer value. To make a vertically integrated marketplace work in Latin America, you must build underlying infrastructure that U.S. companies often take for granted. AI is not just a productivity layer; in complex businesses it must become the operating system that routes work between humans and agents. Co-pilot tools alone are insufficient if employees do not adopt them; agents embedded in workflows are more effective. The best products in emerging markets should solve necessities, not just conveniences, because users are dealing with basic economic constraints. Long-term value creation comes from relentless 1% improvements in user friction, not from chasing short-term growth metrics. Founders must be willing to let go of their old operating style and re-invent themselves as the company enters new phases.

Data Points: Used-car transactions ending in fraud in Latin America: 40% - Carlos cites this as a core reason Kavak exists, contrasting it with the U.S. lemon problem. Used-car market financed in Mexico: 5% - Carlos compares this with the much higher financing penetration in the U.S. Used-car market financed in the U.S.: 90% - Used as a benchmark to show the underdeveloped financing market in Mexico and Latin America. Kavak monthly transactions: around 10,000 - Carlos gives this as current operating scale. Kavak headcount: around 3,500 people - Describes the size of the company at the time of the interview. Kavak employees at peak during scaling: 10,000 employees - Referenced as the company’s size during the 2021-2022 high-growth phase. Growth rate in 2021: 300% - Describes Kavak’s rapid expansion before capital conditions changed. Flat growth period during AI transition: 1 year - Carlos says the company experienced about a year of flat growth while restructuring around AI. Share of customer interactions handled by AI agents: more than 90% - Carlos says agents now handle the majority of customer interactions. Human-to-AI interaction coverage: 90%-95% - He later says AI agents manage roughly this share of user interactions. Personal schooling: 14 schools - Carlos says he attended many schools growing up, shaping his adaptability. Best companies’ value creation after year 15: 97% - Carlos argues that almost all value is created long after a company starts. Share of customers buying a car for the first time: 40% - Carlos says this is a key pride metric showing Kavak’s impact on mobility and class mobility. Mexico City market concentration: 60% of Mexico’s opportunity - Used to justify focusing heavily on Mexico City before expanding. Expansion city ticket-size threshold: below $10,000 did not make sense - Carlos says this was the minimum economics needed for early expansion markets. Most used-car transactions in Latin America: 90% informal - Carlos says the market is overwhelmingly informal, making fraud and unsafe transactions common.

Pivotal Quotes: "When you're building AI, the first thing that you need to build is the brakes of the system." — Carlos Garcia Otati: Explaining why governance, deployment rules, and data ontology must come before broad AI rollout. "We went funnel by funnel, putting agents in front of critical aspects of the business." — Carlos Garcia Otati: Describing Kavak’s shift from employee co-pilots to customer-facing and workflow-embedded AI agents. "I fire myself." — Carlos Garcia Otati: His annual self-assessment process for determining whether he is still the right CEO for the company’s next phase.

Implications: For emerging-market founders, the lesson is to solve painful structural problems, build infrastructure when needed, and use AI as an operating layer, not a gimmick. The winners will be companies that can endure temporary pain to redesign for scale.