This is quite interesting and a very important side effect of AI if we survive long enough for it to matter, but am I reading the layperson prose right that the "Resy Wasn't having it"/"Resy suspends"/"Resy... shut down" language ultimately was just a bog-standard REST API rate limiter, in this case?
That's kinda funny if so, and captures the 2026H1 AI discourse beautifully: so many of the safety stories that broke through the vague rumble of datacenter & IP vitriol were ultimately just apps making extremely elementary mistakes. The guy who went viral because he claimed Railway's AI deleted his whole company, when the actual truth was that he had unencrypted keys on disk and was using snapshots as backups -- and they recovered the snapshot within 24h, anyway!
Similarly, the first sentence in this article becomes false as soon as someone implements a sane, intuitively-devious polling service that checks in irregular periods around 5-30 minutes rather than whatever this was, which was presumably over 60 requests an hour. So I wouldn't consider this solved. Hell, someone will probably read this comment and make it into the next YC batch with a company called NgtvXtrnlTs!
Apologies for distracting from the topic at hand, but as it's always on the mind: I already miss the days where unattended agents were liable to make such fundamental software engineering mistakes...
Original title “This VC used AI to try to snag a restaurant reservation. Resy wasn't having it.” compressed to fit within title limits.
https://archive.today/MJW6M
This is quite interesting and a very important side effect of AI if we survive long enough for it to matter, but am I reading the layperson prose right that the "Resy Wasn't having it"/"Resy suspends"/"Resy... shut down" language ultimately was just a bog-standard REST API rate limiter, in this case?
That's kinda funny if so, and captures the 2026H1 AI discourse beautifully: so many of the safety stories that broke through the vague rumble of datacenter & IP vitriol were ultimately just apps making extremely elementary mistakes. The guy who went viral because he claimed Railway's AI deleted his whole company, when the actual truth was that he had unencrypted keys on disk and was using snapshots as backups -- and they recovered the snapshot within 24h, anyway!
Similarly, the first sentence in this article becomes false as soon as someone implements a sane, intuitively-devious polling service that checks in irregular periods around 5-30 minutes rather than whatever this was, which was presumably over 60 requests an hour. So I wouldn't consider this solved. Hell, someone will probably read this comment and make it into the next YC batch with a company called NgtvXtrnlTs!
Apologies for distracting from the topic at hand, but as it's always on the mind: I already miss the days where unattended agents were liable to make such fundamental software engineering mistakes...