Built from Broken

Real problems we face running AI-powered nonprofit operations... and the real solutions we build. Every fix exists because something failed first. We show the receipts.

Vol. 1

Built from Broken: Vol. 1

The docs existed. The agent didn't read them. 40 minutes of debugging... for a fix on line 47.

ai-agentclaude-code

Vol. 2

Built from Broken: Vol. 2

8 fixes in 6 weeks. Same pipeline. Every fix built on documentation that was already wrong.

ai-agentclaude-codedocumentation

Vol. 3

Built from Broken: Vol. 3

Correct date. Wrong day name. Written with full confidence into an email to a real person.

ai-agentclaude-codehooks

Vol. 4

Built from Broken: Vol. 4

35 minutes of downtime. Two agents. One port. Neither knew the other existed.

ai-agentclaude-codeinfrastructure

Vol. 5

Built from Broken: Vol. 5

I had 21 ideas in a 'don't forget' panel. The panel never reminded me. One sat 4 weeks past due while I worked on the same app daily.

ai-agentclaude-codehooks

Vol. 6

Built from Broken: Vol. 6

Load avg 30 on a 4-core box. Five chats open, five copies of the same helper, roughly 23 GB of RAM... eaten before the work even started.

ai-agentclaude-codemcp

Vol. 7

Built from Broken: Vol. 7

Three weeks of schemas. One end-of-day decision. The framework I built read me back to myself. I teared up.

ai-agentclaude-codevalues

Vol. 8