Applied AI Case Study Library
What actually happened when AI met the real world. Every case researched from primary sources and graded on the strength of its evidence.
HOW IT WORKS
How a case gets made
- IN / OUT
- STEP
- CHECKPOINT
- IN Something happens AI meets the real world
- STEP Primary sources court records · filings · studies
- STEP Trace every claim to the evidence behind it
- CHECKPOINT Grade the conclusion A to C, plus causal label
Weak evidence gets a low grade, not a pass
- OUT Published case open to contributions
Simplified. The full story is on the project site.
The problem
Almost everything written about AI in practice is either a vendor’s success story or a cautionary anecdote. Neither tells you how strong the evidence is, or whether the AI actually caused the outcome being credited to it.
What it is
The Applied AI Case Study Library is an open, evidence-first record of AI in practice. It covers law, work, safety, healthcare, public services, and infrastructure: from training pauses at OpenAI, to an NHS triage trial, to more than two thousand court rulings involving hallucinated filings.
How it’s made
I research and write every case myself, from primary sources only: court records, filings, official reports, the original studies. Each one is heavily vetted before it goes up.
Three rules hold across the library:
- Every claim is traced to the evidence behind it.
- Every conclusion is graded, A to C, on the strength of that evidence, and labeled for how confidently it can be said that AI caused what happened.
- Every revision is tracked in Git, so any change to a case is on the record.
The library is open to contributions.
Why it matters
It’s the evidence side of the AI Operations Management textbook. The book sets out how AI should be run; the library shows what happens when it is, and isn’t.