At the same time, people need to follow a common recipe so to carry forward the knowledge of predecessors. The hard part is sprinkling in the right amount of the learner's own strengths (data and representation) into that recipe.

memory can also be considered as the project "state"

leisure ai coding rate 3/5;
enterprise coding rate 1/5.

I also tried a few popular AI coding integration, so I don’t need to handle infra and basicS. My experience with bolt (.new) and Replit fall short, as the agent is stuck with simple issues (image handle, db errors) which already cost the simple entry plan credit.

AI Coding feels like a wild horse, it can leap ahead or throw you off the saddle.

The next most valuable improvement is engineering structured memory. Humans/Teams have diversity, so should projects.

What’s really annoying?
- unable to memorize the common patterns across projects
- thinking too long
- lying
- inconsistent changes (even reverting previous edit)

The speed from idea to a prototype is so greatly reduced that I cannot wait to implement so many ideas, iterate cheaply and rapidly and pick the winner based on reality rather than cloud in my head.

I tried to learn nextjs and all these newest website framework two years and really struggled with how to format it well and make it responsive etc. simply because I don't know where to start, I can imagine this helps me greatly if I were to start a one-person company one day.