How Enterprise AI Really Gets Deployed
Summary
Sarah Wang and Kimberly Tan interview Decagon co-founders Jesse Zhang and Ashwin Srinivas about shifting 90% of their AI stack to open-source models for better latency and control. They discuss how fine-tuned smaller models outperform frontier models on specific tasks, debunking the idea that apps are merely thin UIs. The episode explores Decagon’s product-led growth, the evolution of forward-deployed engineers, and why enterprise software will endure even with AGI.
Topics discussed
Intro: AI agents as the front door of business
Decagon's origin and the shift to open-source models
Fine-tuning smaller models for cost, speed, and accuracy
Enterprise challenges: evals, governance, and model risk
In-house talent strategy and building internal tooling
Tokenomics: optimizing for latency and unit economics
Application layer vs. model layer: where the moat lies
The trap of forward-deployed engineers in AI products
Productizing customer insights and scaling the core product
AGI impact on jobs and the 'Duet' agent for ops
Decagon's moat: infrastructure and legacy system integration
Sales-led growth and navigating enterprise organizations
Expanding beyond customer support to broader workflows
Product roadmap and the limits of current model capabilities
Hiring trends and the A16Z office culture
International expansion and vertical vs. horizontal markets
The future of CRMs and data records in an agentic world
Founders using AI agents for brainstorming and decision-making
The role of X (Twitter) in shaping AI narratives and hiring
Jevons Paradox: AI increases support volume, not just efficiency
Listen ad-free on Castria