I think this core business model question is happening at all levels in these companies. Each time the model goes in the wrong direction, and I stop it - or I need to go back and reset context and try again - I get charged. The thing is, this is actually a useful and productive way to work sometimes. Like when pairing with another developer, you need to be able to interrupt each other, or even try something and fail.
I don't mind paying per-request, but I can't help but think the daily API revenue graph at Cursor is going up whenever they have released a change that trips up development and forces users to intervene or retry things. And of course that's balanced by churn if users get frustrated and stop or leave. But no product team wants to have that situation.
In the end I think developers will want to pay a fair and predictable price for a product that does a good job most of the time. I don't personally care about switching models, I tend to gravitate towards the one that works best for me. Eventually, I think most coding models will soon be good at most things and the prices will go down. Where will that leave the tool vendors?
I don't mind paying per-request, but I can't help but think the daily API revenue graph at Cursor is going up whenever they have released a change that trips up development and forces users to intervene or retry things. And of course that's balanced by churn if users get frustrated and stop or leave. But no product team wants to have that situation.
In the end I think developers will want to pay a fair and predictable price for a product that does a good job most of the time. I don't personally care about switching models, I tend to gravitate towards the one that works best for me. Eventually, I think most coding models will soon be good at most things and the prices will go down. Where will that leave the tool vendors?