What have I done?
I’m building a dataset.
Late into the night. Night after night. Bleary-eyed, moving from one Rightmove listing to the next. Import the floor plan. Scale it. Draw spaces over the top that link to a database. A pre-formatted A3 sheet populates itself. Export. Done. Next.
Over and over again.
It started simply enough. I was looking through Rightmove for a property I could make a video about — something with a problem that needed a simple but novel fix. After clicking through about fifty, I’d set a few aside and kept scrolling.
Then a thought stopped me.
What if the video wasn’t about one house, but about all of them? If I could find a way to analyse every property in this search – all 276 of them – something might reveal itself. A pattern. A recurring problem that affects not one house but thousands. An insight that could actually be useful at scale.
So I built a process. A sheet for each property: area, percentage of wasted space, remodelling opportunities, value per square metre. I bought a small colour printer. I’m going to need a big wall to pin them all up — arranged by house type, or problem type, or something. I don’t quite know yet. That’s part of the point.
And then, a few nights in, something shifted.
I realised that each sheet, uploaded to an AI, becomes a learning dataset. I tested it on 100 properties in the £200,000–£400,000 range. It can already take a set of requirements and suggest five properties that fit. Pick one, ask it to show how it could be extended, and it will produce a floor plan, a visual, a budget estimate, and suggested next steps.
The floor plan is neatly drawn; it has the bones, but kinda nonsense. But we all know where this is going. It isn’t getting worse. The cost estimate was spot on.
I sat back and looked at what I’d made.
I may have inadvertently started to build a tool to replace me. A robot.
Which brings me to those Silicon Valley types. I’ve spent the last few years quietly baffled by them. All that effort, all that capital, all that disruption — and for what? I’ve thought and written: what was wrong with 1995? Not nostalgically. More — didn’t it all work kind of well enough?
And now, sitting here at midnight with a stack of printed floor plans and an AI that is slowly learning to do my job, I think I finally understand them.
I don’t think they knew. It was just a cool challenge. Build a useful tool. And now it’s too late.
All the best
This Week’s Links:
Air fryer blueberry cheesecake toast
This makes me want to go out and buy an air fryer
Ancillary Dwelling Units in NYC
This initiative looks interesting in that there is a pre-approved plans library with cost estimates for this type of annex building.
Deer’s antlers make great pylons
These images are not real, but the proposal is. Interesting and slightly scary.
Main Image credit: A painful realisation (ChatGPT)
