I build software for brokers and I use these models every day, so this is not a piece against them. It is a piece about what they are good for in a business where a wrong figure in a specification can cost a viewing, and a wrong claim in a brochure can cost rather more.
Where it genuinely helps
Writing the first draft of a listing from a specification sheet and a set of notes, so the broker edits instead of staring at a blank page. Turning a month of recorded activity into a readable owner report, with the numbers already in the right order. Summarising a survey or a refit invoice so the broker can answer a buyer's question in a minute rather than an hour. Translating material for a co-broker in another market without losing the facts.
All four share a property: the facts already exist, and the tool arranges them. That is where these systems earn their keep.
Where it produces confident nonsense
Estimating what a yacht is worth. Predicting when she will sell. Writing a brochure for a yacht it has not been given facts about, which it will do cheerfully, inventing a tender garage along the way. Answering a buyer's technical question without the documents. Anything where the model is asked to supply knowledge rather than arrange it.
The failure is not that the tool is bad. It is that nothing in its output tells you which sentence came from the specification and which came from the average of every brochure ever written. A broker has to check every line, and checking every line of a fluent text takes longer than writing a plain one.
The useful question
Not whether to use AI, but what to feed it. A brokerage that records what actually happens around a listing, the viewings, the feedback, the document requests, the price history, has material a tool can work with. A brokerage that keeps that in a broker's head and a WhatsApp thread does not, and no software will fix that.
That is the unexciting conclusion. The leverage is in the record, not in the model. Build the record first. The report writes itself afterwards, sometimes literally.
Who pays, and for what
Vendors sell these features as time saved. The honest accounting also counts the time spent checking, the mistakes that slip through, and the dependence on a platform whose pricing will change. A tool that drafts an owner report from recorded facts saves hours a month and makes few mistakes. A tool that writes listings from thin air saves one afternoon and costs a reputation. Pay for the first kind.