Work
Two of these handle real customers today. One is in the client's final testing. They work the same way underneath: the AI does the sorting and the looking up, and a person still decides what reaches a customer.
NewmanBands (opens in a new tab), e-commerce (men's wedding rings)
Every ticket used to mean the same manual lookup: read it, pick a canned reply, switch to WooCommerce, find the order, copy the tracking number across. Now a service does that part and leaves a draft in the ticket. The agent reviews and sends.
Lynda Samphire (opens in a new tab), dream interpretation practice
Her app sent her interpretation method, the thing she actually sells, to every visitor's browser, along with the key to her AI account. She asked me to move both somewhere safe. Two to three days from brief to live.
NewmanBands, second project
The same owner came back for a chatbot on his help page. It works out which of about 94 known situations a customer is in, asks his own follow-up questions, then sends the reply he wrote, or hands the conversation to a person with the full history attached. He edits the answers himself in a spreadsheet and uploads it. No developer, no redeploy.
It isn't live yet, so there's no case study and no results to show. I'll write one when it launches.
Four years as a senior full-stack engineer for Finlex, a Frankfurt insurtech, owning features end to end across Ruby, Node.js, Angular, MongoDB and AWS, including a migration of 30,000+ live insurance contracts. Currently building a membership e-commerce store with a paid AI companion.