What if getting genuinely useful at AI took less time than learning to rollerblade?
In the second instalment of Phuel’s AI series, Nathy Gaffney and innovation expert Jake James dig into the practical mechanics of building AI fluency, and make the case that the barrier to entry is far lower than most leaders assume. Jake reframes the whole conversation from the outset: this isn’t an imperative you’re forced into, it’s ground floor access to a shift as significant as the early internet.
The heart of the episode is the 10 hour rule, borrowed from Ethan Mollick’s Co-Intelligence. The logic is elegant: spend 10 hours experimenting with AI in a domain where you’re already the expert. Because you know the subject, you’ll immediately spot when the AI gets something wrong, which takes the guesswork out of figuring out whether a shaky answer is the AI’s limitation or your own prompting.
That leads naturally into hallucinations, and Jake gives one of the clearest explanations going: AI is a mathematical prediction engine dressed up as language, and when it hits the edges of its knowledge, it fills the gap and presents the guess as fact. He also drops a genuinely surprising claim from Mollick: that AI can outperform 90% of people on creativity tasks, while pushing back a little on how far to take that.
The practical core of the episode is the CAGE framework: context, action, guide and evaluate. Jake compares prompting an AI to onboarding a grad or intern. You wouldn’t expect them to nail a task first go without context, so why expect that from AI? The PB&J experiment, where kids give literal instructions for a peanut butter sandwich and a parent follows them exactly, is a properly funny illustration of why vague prompts produce vague or bizarre results.
Jake and Nathy also cover the practical habit of starting fresh conversations rather than arguing with an AI that’s locked into a bad context, and the trick of assigning AI a specific role or persona to pull more specific answers from the edges of its knowledge rather than the generic middle.
The episode closes on a genuinely important caution: research suggesting students who use AI to think with are getting sharper, while those using it to think for them are getting weaker. The line Jake and Nathy keep coming back to: this is about augmenting human intelligence, not replacing it.
This episode is worth your time if you want a practical, jargon-light way to actually start building AI capability instead of just reading about it.
