In a Latent Space podcast interview, Modal CEO Erik Bernhardsson explores the disconnect between how AI workloads actually run and how the market sells capacity in fixed, multi-year contracts. He walks through Modal's approach to pooling , autoscaling workloads, and sourcing GPU capacity in a tight market.
Bernhardsson covers the trade-offs between renting and owning infrastructure, the risk of GPU price declines, and how alternatives to NVIDIA are reshaping the landscape. The conversation also digs into Modal's expansion into agent sandboxes—specialized environments for running autonomous —and the security challenges that brings, including using agents themselves to test defenses.
Topics span developer experience in an agents-first world, the role of forward-deployed engineers, and why Modal has made code review optional in certain workflows.
This is an AI-written summary. The original reporting belongs to @LatentSpacePod. How we write these · Spot a mistake?