using claude today gave me a peek at what the future will look like. the way we get work done, express ourselves, and learn is going to change a lot in the next few years.

the promise of an agent that can automate any work task for you, and help you become a better person is very exciting. but.....its not perfect yet.

claudes introspection ability was lukewarm. i found it tended to misconstrue some information i gave it, and overindex on certain personality traits of mine. for example, it generalized that because i was a PM, i tend to oversimplify systems and skip layers of abstraction.

while this had some truth, claude viewed every problem i encountered in that lense, which was not always helpful.

i was also particularly impressed by claudes memory and ability to traverse through the 32 sessions we had in the last week, identify patterns and identify areas of improvement: 1. too often, i tried to simplify things, and make assumptions. 2. i jump into fixing the code before i really understand what the problem is. 3. i tried getting claude to socratic question me, but sometimes it wasnt' effective. 1. when i just needed a fact like what serialization was, it was harmful, and took longer for me to learn. 4. i should have pushed claude more if its response wasn't to my liking. 1. oftentimes it would ask me 4 questions at once about my understanding. thats too much, and i can tell it that.

principles i will be taking forward:

ensure you provide the right context to claude for quicker loops, and better repsonses - if you make a change to the code, tell claude what kind of change you made. - if you don't understand something, explain what your current mental model is, then claude can correct it and provide more valuable feedback keep a watchful eye, claude can have a tendency to overcomplicate or misconstrue things - when you notice this, push it