one of my ai benchmarks has been if the models / harnesses can automated grocery ordering for me @bot is the first one to solve this well, what it gets right: - a dedicated chat for groceries where i can state preferences i.e "no olives" and it remembers them over time - a browser it can use and i can sign into amazon & costo on - automated tasks to run ordering on a schedule - access to my MCPs (shoutout executor) so it can update my calendar with what to cook and when - having an actual running machine to take over and sign in to sites is so nice what it's missing is collaboration, i.e i want to include my girlfriend on this chat an artifacts system would also be nice, although i can just use executor for that but i need to expand it with storage for mini apps also the way that it uses the browser is very slow - i would've liked to see it have some form of claude workflows style approach where it can parallelize work not as big of a deal as this'll be a mainly background job, but i think codex's chrome plugin still wins out what's maybe most interesting here is how it replaces something like a HelloFresh for me, it's a significantly better experience, cheaper, and i can be more granular with my preferences over time with the fact that it's doing work in the background, if i really wanted i could: - order bulk goods / frozen meat from costco - order fresh produce from whole foods - order misc items from amazon fresh obviously wouldn't want to do that on every order, but it can better order the right items from places whereas i previously wouldn't care to the last thing is on the computer, i'm really wanting a cloud browser that i can connect to my agents over MCP as i get more bots i don't want to have to reauth to each one, same thing with the CLIs that'd be installed on the machines i basically want my browser but running in the cloud, which can be forked or spawned with specific credentials browserbase / kernel may be the answer to that part, but my understanding on them is they're more @bot @bot *understanding on kernel and bb is more infra than consumer the artifacts / memory part is really interesting here imo, my grok bot needs: - knowledge of what's in my pantry - knowledge of the recipes that i like / custom recipes going to try the naive approach of just giving it this info and assuming it's memory can store it, but more to build here for executor actually

— Rhys (@RhysSullivan) https://x.com/RhysSullivan/status/2089338930009374994