Ale Massa, Head of AI at Kavak, detailed the company's radical transformation into an "AI-native" organization, a process that involved deeply embedding AI agents into nearly every aspect of its operations. Before Kavak, Massa ran Opi Analytics, an early machine learning company that he now views as being "10 years ahead of its time." The advent of Transformers and the "ChatGPT moment" provided the breakthrough needed to fully realize his ambitious vision.
Kavak, a used car marketplace in Latin America, had to vertically integrate services like fintech, logistics, and inspections due to a lack of existing infrastructure. This complexity made it an ideal candidate for an agent-driven model. Massa's vision was to rebuild Kavak as if it were 2035 with "GPT-10 level intelligence."
The core of Kavak's strategy is the "agent-per-customer" architecture. Every day, between 100,000 and 200,000 agents are instantiated for specific customers, each with its own virtual machine. These agents remember years of interaction, formulate long-term strategies to maximize customer lifetime value, and aim to keep customers happy across all Kavak's products. Currently, 96% of all customer interactions and 95% of transactions are handled by agents.
Massa highlighted three key decisions for this transformation:
1. **Redesigning the entire company:** Instead of simply giving AI tools to staff, Kavak rebuilt its APIs and systems to enable agents to perform.
2. **Building "superhuman agents":** These agents were designed to outperform Kavak's best human employees across critical metrics like conversion, lifetime value, and customer experience, tackling the company's most challenging problems.
3. **Shifting success metrics:** Kavak moved from measuring transactional outcomes (cars bought/sold) to relational ones, focusing on maximizing the lifetime value of its 10 million customers.
Central to this success are robust evaluation (eval) systems. Massa stated they spend as much time, engineering effort, and tokens on building evals as they do on building agents. This allows them to "move extremely fast" with confidence. Evals measure actual business results, such as customer conversion and satisfaction, rather than superficial KPIs.
The impact of these agents is profound. They are "better sellers," converting customers 2.1 times more effectively than human teams, due to their infinite patience, perfect recall, and instant learning. Kavak's agents also handle complex financial services, approving car loans in under three minutes (compared to months) and offering personalized interest rates. Massa even shared an example of an "AI CEO" managing Kavak's Cuernavaca city operations, achieving 1.5 times profit growth in six weeks by micromanaging daily tasks and coordinating physical workers. Even in the physical world, agents act as "sidekicks" ("El Maik") for mechanics, improving inspection quality, speed, and reducing warranty claims by 20-26%.
The organizational structure has been completely redefined. Kavak launched a "Jedi Academy" to train every employee, from the CEO to mechanics, in AI. This program empowers them to build and collaborate with agents. The company now features flat, highly senior, empowered teams, where agents can be "bosses" and humans design agents. Critically, Kavak avoids the common "human in the loop" pitfall of agents simply escalating issues to human support. Instead, agents call specific APIs for human help, and these human interventions are used as data to continuously train and improve the agents, ensuring a closed feedback loop.
Massa’s advice for other leaders and founders is clear:
* **Embrace top-down transformation:** A clear, vertical vision from leadership is crucial, as superficial adoption of AI won't yield significant results.
* **Measure the ROI of AI:** Focus on "Tier 3 tokens" where the return on investment can be directly measured, rather than unquantifiable general usage.
* **Embrace "creative destruction":** Like previous industrial revolutions (e.g., electricity transforming factory design), AI will lead to new, AI-native companies disrupting incumbents unwilling to fundamentally redesign their operations.
* **Build for the future:** Massa believes this is the "most exciting time in human history" for founders, with unprecedented access to powerful tools. He encourages them to envision an AI-driven future and build solutions for that reality.