A reader wrote the same report twice — once with plain ChatGPT, once with his Teammate Stack. Same person, same AI model, same task. Only one of them sounded like him.

You wanted self-starters. You built a permission structure. No wonder everyone is waiting.

He had a tattoo that said "Call me a baby because dolphins make me cry." He also said he'd pour hot oil on a kitchen robot to keep his job. I met him because I turned down Dave Grohl to hold a sign in Washington Square. Here's what I learned.

The prompt before the prompt is noticing what deserves attention.

I asked my AI chief of staff to read 50 pre-event transcripts and write the FAQ I'd been avoiding for months. It took three minutes.

Reese Witherspoon asked her book club, "do you use AI?" Three million people watched. The internet split — and both sides were right, because they were arguing about the wrong word.

A real concern is not a valid conclusion. An EVP in finance emailed me last week about AI sycophancy. Most people raise that concern as an off-ramp. He made it the curriculum.

There are levels of AI augmentation: Some people use AI to draft emails. Josh used it to run 27 agents on a single problem, from his mobile on the road, with all the context of his laptop at home. Those are not the same thing.

Andrej Karpathy hasn't typed a line of code since December. That sentence has a blank in it — fill it with your word, and something shifts.

When Manu Ginobili raised his hand during a Spurs AI session and said "Could AI...?" — that was the ball game. The answer to almost every "could it" question is yes. The problem is most people don't have a question yet.