Hi everyone,
I am trying to build a decision framework for AI work on Xperience by Kentico projects and would love to stress-test it with the community.
Path 1, in-site AI: we build the feature (visitor-facing assistant, editor helper, AI-powered search). We control prompts, retrieval, costs and logging. The client gets a finished feature, and we get a familiar delivery model.
Path 2, client-side AI: we expose content and operations through an MCP server and skills, and the client’s AI tools consume them. We ship capabilities instead of UI. Questions shift to authentication, permissions, which operations are safe for an agent, and who owns the outcome when the agent does something odd.
My current gut feel:
🧭 Visitor-facing use cases lean toward path 1, because you cannot rely on visitors bringing their own AI
🧭 Editor and marketer workflows are where path 2 gets interesting
🧭 Regulated or data-sensitive clients push everything back toward path 1
Where would you disagree? And has anyone found a good way to scope and price path 2 when the "deliverable" is something the client uses through a tool you do not control?