3 pointsby simonebrunozzi2 hours ago1 comment
  • hiroto_lemon2 hours ago
    Worth noting these "how I use Claude" pieces consistently underweight the eval loop. Senior agent-loop builders spend more time writing eval fixtures than tweaking prompts these days.