18 posts.
Most hackathon projects get no further work once the event ends. The ones that survive share three things the research can name, and all three are decided before and after the day. A one-day format built around them.
A nationally representative survey mapped AI use to real occupations and tasks. At least a fifth of workers use it in 80% of occupations, and fewer than 3% of tasks pass 50% adoption. It also corrects a number I published here.
Microsoft measured 29 factors against reported AI impact across 19,854 people. Culture, manager support and talent practices explained 67% of it. Individual mindset explained 32%.
Berkeley researchers spent eight months inside a company using AI. In the moment people felt momentum. Stepping back, they felt busier. Both were true.
Anthropic sampled a million conversations and found experienced users delegate less, not more. What they do instead, and how it fits the long-chat research.
Stanford payroll data shows a 19% employment gap for young workers in AI-exposed jobs, driven by reduced hiring, not layoffs. Here is the split that explains it.
65% of workers say AI improved their productivity. 89% of executives say it has had no effect on theirs. Both surveys are sound: the gains are real and they're getting absorbed before they reach anything you can measure.
One widely-quoted study says 95% of AI pilots fail. Another says 48% reach production. They disagree because they're measuring something you probably shouldn't be attempting, and the pilot that actually works has a different goal.
For one spring, some companies treated heavy AI usage as proof of a high performer. By summer the bills arrived and the idea collapsed. The mistake underneath it is older than AI, and you're going to be offered it again.
85% of employees say their AI training doesn't help them use AI in their actual role, and it isn't because companies aren't training. Gallup found the thing that moves usage isn't the course at all.