LLMs aren't the golden bullet the article hints at. Sure they are improving, but the cost is not falling. It costs a huge amount to create foundation models, and there will be a point where either we have a breakthrough (ie we move from sequence generation to concept synthesis) or the money runs out.
But regardless the rule of thumb still holds:
If your business idea is simple to do, then you need another plank to make your moat. That could be network effect, access to capital or both.
Patents are there to inhibit capital, because it costs money to challenge a patent (as well as defend)
If your business idea is not simple to implement, then you might have the benefit of time.
AI doesn't really change any of that, it just amplifies the effect. ie, making an amazon clone is simple now, because the tech/infra exists. Amazon had to make that infrastructure first, which was hard.