Modal shifts from developer experience to agent experience

Modal CTO Akshat Bubna discusses the company's strategic shift from traditional developer experience to "agent experience" following a $355M Series C funding round. The article argues that AI agents require tighter infrastructure with fast iteration loops, programmatic control, and isolated sandboxes, as they cannot fill in missing context like human developers.

• Modal is building an AI cloud focused on elastic inference, GPU snapshotting, DeFlash, speculative decoding, and Auto Endpoints.
• The platform utilizes a 17-cloud capacity pool and supports serverless multi-node training for post-training and research workloads.
• Production agents require specialized sandboxes with hard guardrails, as RL rollouts may necessitate up to 100,000 sandboxes.

This evolution addresses the limitations of Kubernetes for bursty AI workloads and provides infrastructure that agents themselves can operate programmatically.