The Shift to Agentic Architecture

Ali Masri, Head of AI at Kavak, argues that the future of organizational structure lies in agentic systems rather than human-centric workflows. Instead of merely integrating LLMs into existing processes, Kavak redesigned its entire infrastructure to support agents that act as long-term stewards for each customer. Every customer interaction spawns a dedicated agent with its own virtual machine, memory of past interactions, and a clear mandate to maximize lifetime value. These agents are not simple chatbots; they are goal-oriented entities that can plan, execute, and learn across days or weeks.

The Primacy of Evaluation (Evals)

Masri emphasizes that the primary bottleneck for AI adoption is not model capability, but the quality of evaluation frameworks. Kavak treats evals as a first-class citizen, dedicating equal engineering time and resources to building evaluation systems as they do to the agents themselves. By measuring business outcomes—such as conversion rates and customer lifetime value—rather than superficial metrics like call duration, the company creates a tight feedback loop. This allows the system to learn from failures: when an agent hits a wall, the data is captured, analyzed, and used to improve the entire fleet of 200,000 daily agents.

Organizational Transformation and the Jedi Academy

To facilitate this transition, Kavak implemented a top-down mandate known as the "Jedi Academy." This program trains everyone from the CEO to mechanics in the art of building and collaborating with AI agents. Masri stresses that this is not optional; it is a fundamental shift in how the company operates. The organizational structure has become flatter and more senior-heavy, with teams either building agents, managing agent skills, or operating in the physical world where human dexterity remains superior. In this new reality, agents often act as "sidekicks" (like the character in Ratatouille) to physical workers, providing real-time guidance that has significantly reduced warranty claims and improved repair quality.

Redefining Leadership and Roles

Perhaps the most radical experiment at Kavak was the deployment of an AI "CEO" for a specific city. Tasked with maximizing profitability, the agent managed inventory, pricing, and personnel, successfully increasing profits by 50% in six weeks. This experiment demonstrated that even high-level strategic roles are susceptible to AI augmentation. Masri argues that leaders must stop treating AI as a hackathon project and instead define a clear, multi-year vision for what their company will look like in an agent-first world.