# Preparing a SaaS Product for AI Without Coupling AI to the Core
Adding AI to an established SaaS platform raises an architectural question: should intelligence live inside the core application or behind a dedicated boundary?
For Humzio, the direction is to keep intelligence capabilities separate from the core HR service while integrating them with the platform's existing authentication and authorization model.
Why the boundary matters
AI workloads can have different scaling, provider and experimentation requirements from transactional HR workflows. Separating them makes it easier to evolve models and infrastructure without turning the core product into an AI-specific system.
At the same time, AI cannot ignore application permissions. An intelligence service should only be able to act on data the requesting user is already allowed to access.
The lesson
AI should inherit the product's security model, not create a second one. A clean service boundary can preserve that rule while allowing AI capabilities to evolve independently.