Adaptation: The Missing Layer Between Apps and Foundation Models

About this episode

Ben Lorica discusses the challenges of enterprise AI adoption and the concept of adaptation with Sudip Roy.

Ben Lorica talks with Sudip Roy (Co-founder & CTO, Adaption Labs) about why enterprise AI adoption stalls in the “last 5%” of reliability — and why waiting for the next frontier model release is usually the wrong bet. They unpack “adaptation” as something broader than post-training, including gradient-free, inference-time techniques that can sit above models to route, combine, and continuously improve behavior. Subscribe to the Gradient Flow Newsletter 📩 https://gradientflow.substac...

People in this episode

Host: Ben Lorica

Guest: Sudip Roy

Topics covered

• enterprise AI
• reliability
• adaptation
• foundation models
• inference-time techniques

Keywords

• enterprise AI
• adaptation
• foundation models
• reliability
• inference-time techniques

Mentioned in this episode

Organizations: Adaption Labs

More episodes of The Data Exchange with Ben Lorica

• AI Agents Are Implemented, Not Adopted · June 11, 2026 · 46 min
• Why Your AI Agent Isn't Ready to Ship (And How to Know When It Is) · June 4, 2026 · 43 min
• Why Foundation Models Haven’t Replaced Classical Machine Learning · May 28, 2026 · 52 min
• When "Garbage In, Garbage Out" Gets It Wrong · May 21, 2026 · 43 min
• As Code Generation Speeds Up, Who Tests the Output? · May 14, 2026 · 44 min
• The Gap Between AI Hype and Enterprise Reality · May 7, 2026 · 56 min