Scene Hook

A friend complained to me that AI writing is just meh. I didn't argue—I used to think so too. That was until I tossed a professional problem that should've taken me two days at AI, and got a usable framework in 15 minutes. That's when it hit me: it's not that my AI sucks, it's that the work I gave it was too shallow.

Where This Idea Came From + Who's Using It

Physicist Alex Lupsasca won the 2024 Breakthrough Prize in physics (the industry calls it the "Oscars of Physics") for his research on black holes. Last year, he casually tossed a newly published paper problem to the latest GPT—a problem he had spent a very long time figuring out back in the day. GPT reproduced it in 30 minutes. Even earlier, he asked AI to help with a derivation that should have taken several days; it was done in 11 minutes. The funny thing is, many people online thought the new model was "just okay" because using it to write emails felt about the same as the old version. Alex put it bluntly: GPT-3 could already write emails; how good can an email get? The real leap is in the deep end of your profession, not the shallows.

Replication Cost Today

Money: ChatGPT Plus or Claude Pro subscription at $20/month. No extra cost if you already have it. Time: 30 minutes for a trial run. Technical barrier: Zero code, just know how to type. First step: Open your AI chat box, throw in a hard problem from your profession that usually takes over half a day, exactly as it is. No prompt templates needed; just ask it like you would a colleague.

Advice by Stage

Just starting out: If you're still figuring out your direction, using AI for simple tasks is totally fine. No rush to hit the deep end right now. Solidify your expertise first; there will be plenty of opportunities later. Have 1-2 clients: If you already have steady deliverables, I'd suggest picking the most brain-draining part of a project and letting AI tackle it directly. Don't worry if the result is rough; just see how far it can get. Scaling up: If your team is running multiple projects, I'd suggest picking the most time-consuming brainwork in each project, having different people attack it with AI, and comparing the results. Find out which step AI can save you massive chunks of time on—that's your leverage point.