Two clients showed up with the same UX problem: visitors land on a complex site and can’t find the right path forward. One project ended up shipping a GPT-powered search block. The other shipped a deterministic, persona-driven block on rails, no AI at all. Both worked. Neither was wrong.
This talk walks through both production engagements side by side. The AI search block, built for a community college’s 643-course catalog, comes with the production-hardening list nobody mentions in the demos: rate limits, kill switches, prompt-injection mitigations, token ceilings, and cost predictability. The persona-by-goal Mosaic block, built for an industrial services company after their legal team rejected an AI chatbot SOW, comes with a different lesson: deterministic UX, full editorial control, zero model cost.
Attendees leave with a framework for when an AI block belongs in a project and when a deterministic block does the same job better, plus a reusable persona-by-goal pattern they can adapt on their next engagement.
