Working with LLMs°hard word is a speed-run°hard word through 50 years of software engineering.
A few weeks ago, I found myself wanting .pyh°hard word header files°hard word. Just the function°hard word signatures°hard word (and a one-line°hard word comment°hard word). No need to waste the valuable°hard word processing space on implementation°hard word details that it shouldn't need to know or change.
----
Today, I find myself wanting code-file°hard word tiers°hard word. The base layer (database models, LLM°hard word APIs°hard word, etc°hard word.) get uploaded°hard word to Claude°hard word as "project knowledge". The higher level layers°hard word (HTML°hard word templates°hard word) only get uploaded°hard word within°hard word individual°hard word chats°hard word.
💬 the two hardest problems in Computer Science are cache°hard word invalidation°hard word, naming°hard word things, and off-by-one°hard word errors.
Because the API°hard word tools aren't there to keep everything in-sync°hard word automatically°hard word. And ... while I could keep track, I would prefer not to.
----
The "halfway°hard word" solution is a script°hard word that moves a current version°hard word of all the "uploading°hard word" files°hard word to a single directory°hard word. Then°hard word, I use that to create a new project. 💡 projects are cheap. I could upload°hard word a hundred°hard word versions°hard word of the 100KB°hard word of code°hard word files°hard word and nobody would care. 🔥 and, thanks to Claude°hard word, instead of 80% of the work being write the script°hard word, 80% of the work is figure out what the script°hard word should do
This speaks to the iron law of optimization°hard word. Before, it was 20% of X planning, and 80% of X coding°hard word. Now, it is 20% of X planning, and 4% of X coding°hard word. The machine has made coding°hard word 20 times faster in this example, and made the overall°hard word project 4 times faster.
No additional°hard word amount of "faster coding°hard word" can improve this more than 20%. ⚔️ well, actually, in the future, the machine will also be able to do planning