Joris

Joris helps people buy houses in Newcastle. He started with no site, no rankings and established agents to compete with. The first job was to look established and be found.

The second job showed up once the leads did. A buyer's agent runs on admin: a property report for a client means generating it on a data service, pulling the details together, finding the client's record, writing the email. Enquiries arrive as Instagram messages, Facebook messages and website forms, and each one needs reading, judging and logging before anyone calls back. Footage from inspections piles up on a phone, unposted. All of it was one person, after hours.

Joris already pays for Claude and ChatGPT. Rather than give him another app to log into, we built a custom MCP server he installed in both, so he asks in a normal conversation and our tools do the work behind it.

"Send Sarah the report on 12 Smith Street" is now one sentence. The MCP generates the report on the property data service, fetches every detail back, looks Sarah up in HubSpot, builds an HTML report and emails it to her, then tells him it is done. Because we host the MCP, adding a new ability is a matter of switching a tool on: no update for him to install, no new interface to learn.

A worker receives every message from his social accounts and every submission from the website. Claude reads each one, qualifies it (a buyer with a budget and a suburb is not the same as an agent selling leads), creates or updates the contact in HubSpot, places it in the right sales pipeline stage and creates the to-do for Joris, with the context already written. He opens HubSpot to a list of people worth calling, not an inbox to triage.

The next system, in progress: Joris dumps the day's inspection footage into a folder. The AI watches it, picks the moments, cuts them into short edits, writes the captions and posts them to his social accounts on a schedule. Same rule as everything else we build for him: he approves the first ones, and the system earns its autonomy from there.