SERHANT.'s engineering team shares how they scaled their AI platform S.MPLE from 200 internal pilot users to 900+ real estate agents using Next.js, Vercel, and the Vercel AI SDK. The key strategy was avoiding vendor lock-in by using the AI SDK as an abstraction layer to orchestrate multiple model providers — Claude Sonnet for complex analysis, Claude Haiku for lightweight tasks, OpenAI for conversational features, and Gemini for image and browser automation. Vercel's Fluid Compute handled scaling automatically without infrastructure changes, while AI Gateway provided unified visibility across all apps and prototypes. The team is now moving from linear single-action workflows toward conversational multi-agent experiences.
Table of contents
Impact at a glanceUsing multiple models to balance cost, speed, and complexityWorry-free scale: Adding users and assetsFuture-proofing an unpredictable landscapeSort: