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

Modal is a cloud platform that has evolved from a developer-focused runtime into specialized infrastructure designed for the unique demands of AI agents. While traditional cloud providers and Kubernetes were built for human-managed web applications, they struggle with the bursty, compute-heavy, and highly specific requirements of modern AI workloads. Modal addresses these challenges by providing a serverless environment that allows agents to programmatically spin up isolated sandboxes, manage GPU resources, and execute code with fast iteration loops.

The platform distinguishes itself by moving beyond traditional infrastructure assumptions. It offers features like elastic inference for custom models, GPU snapshotting to handle cold starts, and the ability to scale across a capacity pool of 17 different cloud providers. By using a decorator-based approach, Modal allows infrastructure requirements to be defined directly within the code, which enables AI agents to self-provision and manage their own environments without needing to navigate complex, human-centric configuration files or dashboards.

This shift toward agent experience is critical because, unlike human developers, agents require tight, reliable feedback loops to operate effectively. Whether it is running 100,000 sandboxes for reinforcement learning rollouts or providing hard guardrails for production agents, Modal focuses on the primitives that AI applications actually need. By prioritizing observability and programmatic control, the platform enables researchers and developers to build, test, and deploy AI systems that can autonomously handle complex tasks, effectively turning infrastructure into a tool that agents themselves can operate.