it does drive me sort of nuts how people talk about AI agents, and a lot of new AI products in general, as if they're deploying into a static environment. people react when robots start showing up!

ha, now i'm imagining a gui/api hybrid gradually refined away from pure human design principles into something... else

totally, though parties involved have different views. in a world where everyone is begging OpenAI for integration, ChatGPT becomes *much* more capable. Not really unique to AI — Facebook felt this way in the 2010s, any big plat does — but maybe Agent Mode offers a way to force the issue a little?

gotcha — anyway, the presence of the wheel isn't important to the argument about agent stuff, just something that you can observe that's downstream from deeper integration

obviously you don't really get a pure version of either long-term, but it's not clear where between them you end up? in 1 the other parties obviously don't just sit still, they're either persuaded to move more toward 2 or become antagonistic. but 2 is scary for many firms *now*

the broader argument is just that, if the goal is seamless commerce via chat, there are two extremes for how to approach it: 1) built a model so good at using a computer that you can just make it do stuff for you and 2) shrinking the challenge by working together, opening apis, whatever

I think I'm responding to something I don't quite get but: Agent Mode can buy things for me from Walmart. ChatGPT can't. Agent is clearly more capable in that sense, but its process is less direct and more complicated than, for example, a situation in which Walmart partners with OAI

oh, just that there's no vestigial wheel. but yes: waymo talking to regular interfaces makes sense. the contrast is meant to be with a theoretical humanoid robot sitting in the driver's seat instead

you mean like: agent mode immediately works better for ecommerce than the non-solution in sandboxed gpt?

sorry, don't totally follow

I think it's mostly because they were retrofitting the Jags, the Zeekrs aren't like that. And I don't mean to imply they're talking to each other — I'm referring there to the interface between computer and car (via robot/wheel or cursor/web GUI)

or maybe the purple line bsky.app/profile/toda...

The success of Ring is just an incredibly American story: Individualism expressed through consumer habits resulting in the voluntary creation of a national surveillance network. Add home security cameras to the list with guns, cars

yeah, totally see that. my (incorrect) early intuition was that they'd be ok at automating search work, for lack of a better term, chasing down information buried in a pile of links

yeah, I mean Search is pretty grimy these days, but I'm not sure a results page reformatted as an eye-glazing "research report" provides a whole lot more value

Most of the time the sequence is, like, "hey, make me a report that I will not read," which I understand is a major work type in corporate America, but is also a pretty big comedown from the "how does it do that??" coverage this stuff got early on (like this www.nytimes.com/2025/03/04/o...)

Bringing this up mainly because they were held up as an early agentic success. But even with new models, Google DR is returning me bogus, annoyingly verbose reports that extrapolate from out-of-context Reddit comments, fail to answer basic questions, etc. Anyone else having similar experiences?