

Modal and Scaling AI Inference with Erik Bernhardsson
53 snips Jul 31, 2025
Erik Bernhardsson, Founder and CEO of Modal and former Spotify architect, dives into the world of AI workloads and serverless computing. He shares his inspiration for Modal and discusses the significant challenges in machine learning development, especially around slow feedback loops. The conversation highlights the platform's flexibility for diverse AI applications, from music to biotech, and the intricacies of optimizing AI performance for low latency. They also touch on capacity planning for generative AI, resource pooling, and innovations in distributed training.
AI Snips
Chapters
Transcript
Episode notes
Erik's Founding Story
- Erik Bernhardsson founded Modal because he personally faced tooling gaps during his seven years at Spotify building music recommendation systems.
- His experience led him to build biotech workflow tools, realizing a need for better AI infrastructure.
Fast Feedback Drives Productivity
- Fast feedback loops directly correlate with developer productivity in AI and ML workloads.
- Current cloud tooling introduces friction unlike front-end development's sub-second feedback loops.
Modal Platform Overview
- Modal offers a multi-tenant cloud platform with GPU scaling and zero installation using a Python SDK.
- It enables rapid iteration and autoscaling, helping customers run ML and Gen AI workloads efficiently.