Y Combinator Startup Podcast · Startups & Venture · November 2026
Glyman had argued that for many tasks, like accounting, models are already smart enough. He claims that cheaper intelligence used to drive usage up by more than the price cut, and that this stopped in Ramp's spending data recently, though he later allows it may be a local maximum.
…115, the returns to that are going to be incredible. But there's this question of what happens when you go from like 120 to 150 for tasks that don't require it. Like put differently, you know, it might be that accounting is solvable with 120 words. Like we're good. We're good on accounting. I have
friends who, you know, fired their CPA and Claude does their taxes now.
CPA Claude. And what I would say really directly that no one has talked about this, but it's very clear in the data, which is over the last two months, Jebin's paradox broke.
Fascinating.
Like it used to be for that period of time, these are made up numbers. Let's say you discount the price of intelligence by 10%. It becomes 10% more affordable. You would see the increase in consumption of quantity of intelligence used more than offset.
Like railroads came down in price and they spiked up the usage of railroads.