back
user profile
dima853
6karma·12submissions·August 5, 2026
recent activity (12 total)
comment
The "centaur" point is spot on. The only AI use cases that actually work are when a skilled dev uses it like an autocomplete on steroids. The 0% success rate in enterprise AI is wild though.…
comment
How do you handle partial success in the benchmark? If an agent picks the wrong tool mid-chain but fixes it with backtracking, does it still count as a full success?