Whether it’s 3% or 4% I think the growth comes from people and businesses producing more with the same headcount. Early movers will capture most of it before it becomes the baseline, so figure out right now where AI 10x’s your own output.

Honestly, this narrative makes sense. There may be an incentive. But “incentive exists” doesn’t mean “crisis is fake”. I haven’t seen Eisman say anything publicly about recent AI incidents and/or safety/alignment research.

The gap is so wide it blows my mind. The average person opens up ChatGPT a few times a month for a "Google search" or to make a funny image. Meanwhile, we're having national debates about banning superintelligence and human extinction.

Dario Amodei talked about what a one-person billion-dollar company could look like. He didn't give an official checklist, but I turned the examples from his answer into three filters to choose a business I could realistically build. → Does the business deploy its own capital? Dario's example was a proprietary trading firm. A real estate flipping business or used car dealership follows a similar pattern: use your money to buy something, improve or trade it, and sell it for more. You may not need thousands of customers, but you do need capital, expertise, and the ability to absorb losses. I ruled this route out because I wanted something someone could start without risking a large amount of their own money. → Can software deliver the result? Claude Code makes building useful software much more accessible. But I could still build myself a full-time job if every customer needs custom onboarding, a sales call, and ongoing help. I wanted a result the software could deliver repeatedly. → Can sales and routine support be automated without making the customer experience worse? This narrowed the options the most. The customer needs to understand what they're buying, get started easily, and receive value without a consultation. A file converter fits that pattern. A product that requires me to explain a different implementation to every customer probably doesn't. I landed on Agent Report Card: software that tests customer support agents and produces a report showing where they fail. An agency can connect its agent, run the tests, and get a deliverable it can share with its client. Routine questions can go through AI support, while refunds and account deletion still come to me. That's the kind of workload I was trying to design for one person.

Higgsfield's NEW API finally allows you to pay per generation rather than a monthly rate with credits. It isn't automatically cheaper though... It's all about finding the breakeven point.