SERHANT.'s playbook for rapid AI iteration
By Steven Van ·
SERHANT. used Vercel's AI SDK and AI Gateway to scale its S.MPLE platform from 200 pilot agents to more than 900 without rewriting its backend.
SERHANT., a real estate and media company, grew its AI platform S.MPLE from a 200-agent pilot to more than 900 real estate agents without a backend rebuild, according to a case study from Vercel. S.MPLE started as a Next.js progressive web app on Vercel, a foundation that let the team later add a React Native iOS app while keeping the same API layer.
To keep pace with changing AI models, SERHANT.'s engineering team adopted Vercel's AI SDK and AI Gateway. The AI SDK abstracts away differences between model providers, so switching or adding a model becomes a configuration change instead of a rewrite. The team now assigns models by task:
- Claude Sonnet for comparative market analysis and other accuracy-critical, structured-data reasoning
- Claude Haiku for lightweight intent and field-filling tasks where speed matters
- OpenAI models for conversational voice and general chat
- Gemini for image generation, browser automation and computer-use workflows
AI Gateway gives the team a single view of token usage and API calls across prototypes and production apps, according to Technical Director Greg Parsons. Fluid compute handled the jump from pilot to production scale without infrastructure changes, VP of Engineering Jeremy Bunting said. SERHANT. is now building toward conversational agents that let users steer and correct workflows mid-run and combine multiple task "recipes" into one request.
