Why AI Infrastructure must evolve for Agent Experience — Akshat Bubna, Modal CTO

Modal serves as a specialized cloud platform engineered for AI applications, including inference, training, and sandbox workloads. Originally conceived as a serverless runtime to simplify complex Kubernetes-based workflows, the platform has evolved to address the unique demands of AI-driven compute, such as elastic autoscaling and specialized hardware acceleration. The core philosophy emphasizes co-locating infrastructure requirements directly with code, facilitating a seamless transition from developer-centric workflows to agent-centric ones. Technical advancements like block-based speculative decoding and private overlay networking enable frontier-level performance and efficient distributed training. By abstracting away infrastructure complexity and managing capacity across multiple cloud providers, the platform allows companies to deploy production-grade AI models and autonomous agents without the overhead of traditional container management, effectively supporting the industry shift toward compute-intensive, agent-based architectures.