20 posts.
In a randomized trial, students using an AI tutor scored 127% better on practice problems and no better on the exam. The ones using plain ChatGPT scored 17% worse. Here is what to do differently.
77% of workers scrutinize a colleague's work more closely when they know AI was involved. The load-bearing words are 'when they know.' A review method that doesn't depend on being told.
I gave my team AI training, tools and encouragement. Someone still built the useful thing on their own time, because I never said it counted as work.
AI presentation tools give you clean slides and an empty argument. Why default AI slide decks fail, and the assertion-evidence fix that takes ten minutes.
Typing "no, that's wrong" is the weakest move you have. Models average 39% worse across a conversation than in one shot. Here is what to do instead.
One in four executives says an AI error has already reached outsiders or the board, and 84% are confident in AI output without human review. Here's the half-hour review that turns one mistake into a fix instead of a rumour.
39% of workers think AI is eroding their skills. The research points at something more precise than usage volume: how much you trust the tool predicts how hard you think. Here's the routine that protects the skills worth keeping.
Most role redesigns fail because they start with the whole job. Google's data says AI touches about a fifth of the tasks in a typical role, so that fifth is what you rewrite. Here's the procedure, and the before-and-after description it produces.
You don't build an agent, you configure one, and it comes down to four decisions, none of which involve programming. The fourth is the one nobody mentions and the one that keeps you out of trouble.
A short AI workshop where everyone builds one real thing, the agenda, why each block is that long, and the fifteen minutes of scope-cutting that decides whether it works.