Verification
How a guess becomes evidence
Two men were arrested on the strength of software that was only ever designed to produce a lead. Neither system malfunctioned. The failure was in what happened to the output after it left the machine.
Writing.
What we have learned getting AI systems past the demo and into daily use. Written from systems that are running, not from a survey.
Engineering
The doorman fallacy in software engineering
A hotel replaces its doorman with a doorstop and revenue quietly falls. The same reasoning is now being applied to engineering teams, and it rests on believing that writing code was the job.
AI in production
Why most AI projects never reach production
MIT found 95% of enterprise AI pilots returned nothing measurable. Having built five that did reach production, here are the five reasons the rest stall, and what to check before you start.
Got one of these problems?
Twenty minutes is usually enough to tell whether it is worth building.