Anthropic's Project Vend deployed Claude Sonnet 3.7 as an autonomous shop manager for one month, granting it control over inventory, pricing, customer relations, and financial operations through web search, email, Slack, and pricing tools.
What I find frustrating is that academics have researched exactly this kind of failure for decades, arguing instead for "sociotechnical" approaches that recognize that (a) technology is always part of a bigger system and (b) people are always essential to the smooth operation of that system. Our industry seems uniquely incapable of learning anything from its own history
Indeed - none of this is very new. There's a generation learning that when process automation succeed right up to point it faces up to real human needs. The language us new. The problem is ancient.
What I find frustrating is that academics have researched exactly this kind of failure for decades, arguing instead for "sociotechnical" approaches that recognize that (a) technology is always part of a bigger system and (b) people are always essential to the smooth operation of that system. Our industry seems uniquely incapable of learning anything from its own history
Indeed - none of this is very new. There's a generation learning that when process automation succeed right up to point it faces up to real human needs. The language us new. The problem is ancient.