I wonder if the repeated choice to change or stop serving models in ways that break third party agents and applications is a way to prevent others from capturing the value the models can generate?
I also wonder what it would take for the big labs to invest a few millions in basic marketing and product development or UI/UX work, where a lot of gaps seem bafflingly simple to fix yet linger for many many months.
Based on my own experience earlier this year helping my own employer determine what AI products to subscribe to, this is a significant barrier to adoption. I would put together presentations for the execs, get asked simple questions, and have to say things like, “Oh, you can’t find that info on their sales and product pages, or their FAQs, those are all over a year out of date. Instead I’ve been reading Sam Altman’s Twitter feed, a few tech bloggers who seem to be fairly reliable, and user forum posts where I filter out responses more than a month old, then piecing it together myself. Oh, also, this bullet point is wrong, we made the slide Thursday and they completely changed that on Monday.”
To Anthropic’s credit they seem to have backtracked for now (re. Windsurf). I think over time it’ll more be that foundation models end up building more first party apps, putting them directly in competition with app developers. I don’t think API revenues can justify even current day valuations for frontier AI labs so they have to go there.
Thanks for the analysis! Very much agreed.
I wonder if the repeated choice to change or stop serving models in ways that break third party agents and applications is a way to prevent others from capturing the value the models can generate?
I also wonder what it would take for the big labs to invest a few millions in basic marketing and product development or UI/UX work, where a lot of gaps seem bafflingly simple to fix yet linger for many many months.
Based on my own experience earlier this year helping my own employer determine what AI products to subscribe to, this is a significant barrier to adoption. I would put together presentations for the execs, get asked simple questions, and have to say things like, “Oh, you can’t find that info on their sales and product pages, or their FAQs, those are all over a year out of date. Instead I’ve been reading Sam Altman’s Twitter feed, a few tech bloggers who seem to be fairly reliable, and user forum posts where I filter out responses more than a month old, then piecing it together myself. Oh, also, this bullet point is wrong, we made the slide Thursday and they completely changed that on Monday.”
To Anthropic’s credit they seem to have backtracked for now (re. Windsurf). I think over time it’ll more be that foundation models end up building more first party apps, putting them directly in competition with app developers. I don’t think API revenues can justify even current day valuations for frontier AI labs so they have to go there.