New e-commerce agent benchmark this week: 107 real procurement and fulfillment tasks. Best model on the board clears about 62%.
The other 38% is emails that never got labeled and shipments that never left.
i find myself building apps that i want just because i want to remove a silly dialog in the current app that i use. google remote desktop was asking me to use chrome every time and i hated that, so i built CTRL. way faster, more features, and it's mine so i can change it however
Fable doesn't seem like as good of a model as everyone initially thought it would be. It's fine, but seems to offset the longer runs and better abilities with slow performance and high costs, which is a low value proposition.
