Nothing Failed
Every failure in here reported success.
The benchmark that didn't exist arrived in the same confident sentence as the real numbers. The demo that leaked its own key rendered fine. The jobs that never delivered came back with checkmarks. The build went green over an app that didn't run.
Thirteen essays about running AI tools in real work, and the habit of not trusting the indicator.
Thirteen essays ยท three parts
01The Benchmark That Wasn't
A model built a business case on ROI and MTTR figures sourced to benchmarks that didn't exist.
02The Demo That Would Have Leaked Its Own Key
A booth tool one copy-paste from broadcasting its API key to every visitor.
03return<
A missing space, twelve times, hiding behind a green build.
04The Docs Describe a Different Runtime
The documented platform and the one that actually runs are two different documents.
05Fifty Skills and What They Actually Taught Me
The skills I captured from mistakes compounded. The ones I planned didn't.
06Dozens, Not Hundreds
The folder holds 244. The practice holds dozens. Only one number survives being checked.
07It Has Never Once Told Me No
A tool that agrees with everything is only answering.
08When the Stack Falls Over
Three silent failures in one month. Zero alerts. I checked the action and assumed the outcome.
09Not a Demo
A tool used live on an iPad by someone who had stopped watching it and started using it.
10No Pedigree
The work is the target. The credential was only ever the proxy.
11Same Instinct, Different Substrate
Reading a bar, a QC line, a system: one skill that never got the memo to specialize.
12Construction, Not Recovery
When there's no version worth returning to, the honest move is to build forward.
13Prosthetics, Not Amplifiers
The stack is there to remember what I can't.