It seems that current advantages would compound with AI. I.e., if I am making a SaaS for Popsicle stick makers today, why I am disadvantaged with AI vs a new competitor in the space? I guess the hypothesis is the Popsicle stick maker will vibe code all of the software that they need instead. For that, we need significantly better AI than we have today - perhaps something like a 1000X improvement. Basically, this is a world in which non-technical grandparents can vibe code anything that they want. This means, it understands what you want without you being able to articulate it well in the first place.
That's not a 1000X improvement in current AI. That's more like a 2x ~ 5x improvement in current AI, which is measured in months.
So, within months your prompt can simply be "increase NPV" and machines will do the rest? If not, what prompt will work perfectly in your estimation by the end of the year?
“Analyze these business requirements and create a software solution for the problems you identified”.
Right now it can get part way there but quickly falls flat.
In 12-24 months? It’ll be able to audit itself and determine how to fix issues as they come up, mid-stream. That’s (all of) what a human dev does.
How detailed are the business requirements however? "Increase NPV" is certainly a business requirement, albeit a very abstract one. "Add a checkbox to this form" is another, far more concrete business requirement.
Good AI asks clarification questions. Codex plan mode is already getting there-ish.
I suppose it is really only possible to know how close something is until it arrives. When I type in "Maximize NPV" into Codex / Claude Code, I feel like it is incredibly far away from full autonomous capability. I guess we will see.