A record of what breaks when software is built by AI.

Every case here happened to a real business. They are filed not by date but by cause — because the same failures keep arriving in the same shapes, and a cause you can name is a cause you can look for.

If your app already holds a real customer's data, this register is about you. If it does not yet, read it as a warning rather than a diagnosis.

Cases on file

0 failure classes
Primary sources on every entry
Determinations attached

Why this register exists

Building software without engineers is now normal. What has not caught up is any shared record of how it goes wrong — so every founder discovers the same failures privately, expensively, and usually after their first real customer has arrived.

These are not exotic accidents. They repeat, in the same shapes, for reasons that can be stated plainly. Most are cheap to prevent and expensive to fix, and the gap between those two numbers is where businesses die.

How cases are handled

01Evidence, not anecdote. Every case is filed against a primary source — the original disclosure, report, or the builder's own account.
02The subject is never named. Companies and platforms already reported publicly are named. Individuals are not, ever.
03A case has a determination. Not a summary — a stated cause, in one sentence. If it cannot be written that way, it is not yet a case.
04Losses are stated as reported. Where a figure was never established, the entry says so rather than estimating.

Which of these describes your system?

That is the one question these cases cannot answer for you. Bring it to a free 30-minute call: we read what you actually built and tell you, in plain language, which failure classes are present, which are likely, and which you can safely disregard.

Book a call