Jagged Frontier: Home Renovations

As I use AI to help me learn about building renovations, I'm getting some great front row seat failures of its ability to regularly reason things.

For example I need to fashion a window sill pan on site cause my house is structural brick with granite external sill. The box stores only stock liquid sill pans for this situation, but they only last 25-50 years and are near impossible to remove. Im trying to renovate in a way that the house can survive another 150yrs no problem. So copper sill is the choice here.

I learned all this doing research with AI BUT it missed one crucial idea. It had plans to cut the copper sill and solder that corner together.

CleanShot 2026-05-13 at 11.12.14@2x.png

This creates a failure point.

The right move here is folding dog eared corners that have no cuts. (Less tools, stronger).

CleanShot 2026-05-13 at 11.16.43@2x.png

This is a wild small, precise detail that given the code ability of Opus 4.7 I'm surprised it missed (and missed every time, I tried a few different sessions to check). Yet here we are, at the jagged frontier.

Why the heck am I considering going to all this trouble to add a window will pan?

Usually a window install as 2 layers of water management (1) nail flange + tape + house wrap (WRB) and a (2) perimeter sealant.

Since this is a structural brick building we only get perimeter sealant. Should that ever fail, water will end up on the lower wood buck.

Other asides:

Yes, I confirmed from primary sources that dog-eared metal corners is a decent idea and is done. Check out "The Window Experts" showing you how here.

I even tried to use OpenAI's new image model to produce these drawings but it was taking too long to prompt them to the right output (perhaps my prompt skill lacking, dunno).

Im tempted to classify these failures in a bucket of "reasoning about physical systems". But I'm not sure that is the right mental model for this failure mode.