i think it's because they've fine-tuned an old opus 4.6 base model more and more with code
to the point where it's really great at coding, but lost its knowledge on how to write well!
if instead opus 5 was a full fresh training run, i don't think it'd be this trash at writing.
really great coding perf, but now the weights are adapted to so much code fine tuning that it writes like a weird engineer that's trying to sound smart.
Hey! Really great question. My plan is to charge a ~$2 a month subscription if you'd like to analyze more than the last 6 months’ worth of your data (I'll also allow a one-time lifetime license option!).
And I can let you use it on this 6-month window for free because it costs me nothing per user! Your own device does all the processing :D
all write actions are human-in-the-loop by default. these actions are sent to a user's inbox where they can approve/deny requests and also view the lineage of A2A conversations that led to the action being proposed. this has been super helpful for debugging agent teams without taking on any risk.
but if you want, you can configure 'fast mode' where you let your agents take all actions (read/write) without human approval.
to the point where it's really great at coding, but lost its knowledge on how to write well!
if instead opus 5 was a full fresh training run, i don't think it'd be this trash at writing.
really great coding perf, but now the weights are adapted to so much code fine tuning that it writes like a weird engineer that's trying to sound smart.